Search

Image denoising is a vital image processing phase aiming to improve the quality of images and to make them more informative. In this paper, we propose a blind denoising approach for removing the outliers (impulsive disturbances) from digital images, by combining the random sample consensus (RANSAC) and compressive sensing (CS) principles. The proposed approach exploits the fact that images are highly concentrated in the domain of two-dimensional discrete cosine transform (2D- DCT). The sparsity (high concentration) in the transform domain is used in both detection and reconstruction of pixels affected by high disturbances. The image pixels not affected by the noise are found using the RANSAC-based methodology and they are further used as available measurements in the CS reconstruction. The affected pixels are considered unavailable and they are recovered by the CS procedure. The presented approach does not require any disturbance-related assumptions regarding the statistical behavior of the noise or about the range of its values. The theory is verified on examples with 55 images. The comparative analysis against several state-of-the-art methods, done with full-reference and no-reference quality metrics, suggests that the proposed method can be used as an efficient tool for image denoising.

To integrate the effective information and improve the quality of multi-source images, many spatial or transform domain-based image fusion methods have been proposed in the field of information fusion. The key purpose of multi-focus image fusion is to integrate the focused pixels and remove redundant information of each source image. Theoretically, if the focused pixels and complementary information of different images are detected completely, the fusion image with best quality can be obtained. For this goal, we propose a focus-pixel estimation and optimization based multi-focus image fusion framework in this paper. Because the focused pixels of an image are in the same depth of field (DOF), we propose a multi-scale focus-measure algorithm for the focused pixels matting to integrate the focused region firstly. Then, the boundaries of focused and defocused regions are obtained accurately by the proposed optimizing strategy. And the boundaries are also fused to reduce the influence of insufficient boundary precision. The experimental results demonstrate that the proposed method outperforms some previous typical methods in both objective evaluations and visual perception.

Digital image has a significant importance in many fields in human life such as, in medicine, photography, biology, astronomy, industry and defense. Thus, it attracts the attention of large number of researchers, among them those interested in preserving the image features from any factors that may reduce the image quality. One of these factors is the noise. Thus far, solving this noise problem remains a challenge point for the researchers in this field, a huge number of image denoising techniques have been introduced in order to remove the noise with taking care of the image features (edges, sharpness). However, besides that, the findings proved to be inconclusive yet. From this point, the current paper aims to introduce a new denoising method for removing salt & pepper noise from the digital image through spatial way. This denoising method exploits the relationship between pixel’s values when the image changes color. Which gives ordered sequences of values in the four directions, horizontal, vertical and diagonals of the window. The proposed method relays on this concept to change the corrupted pixel, by using the neighbors in the window to extracts the truest value (subjects to this sequence) of the treated pixel. This method has been proven to be simple, effective and performing well comparing with the existing restoration methods with low computational cost.

We proposed an improved phase-field model with local information for image segmentation. In our approach, a new double well potential function containing local contributions is taken into account. The advantages of the new scheme are in two folders. Firstly, the affection of noise could be reduced in image segmentation with intensity inhomogeneity. Secondly, a more stable boundary evolution could be achieved compared with the traditional way. Segmentation results for several different types of images are reported. We compare the segmentation results with the classical model for two-dimensional images. In the segmentation of three-dimensional magnetic resonance brain images, the accuracy comparisons are presented by the Dice and Jaccard similarity coefficients. Compared with the classical model, our model has higher segmentation accuracy and better stability of boundary evolution.

We proposes a multiple fractal dimensions (MFD) method for robust object description. MFD is an effective feature extraction approach, which is first calculated based on a phase angle quantization method to categorize the points of the input image. And then fractal dimensions are calculated to describe the distribution of feature pattern characterized as the intrinsic property of the general objects, i.e., land scene, face and pedestrian. We theoretically proven that our MFD is shown to be invariant to local variations, i.e., Bi-Lipschitz, which is a desirable characteristic for objects, such as land-scene images, face and pedestrian due to the existence of scale variations, local variations and illumination variations in those images. The proposed method is extensively evaluated on land-use scene recognition, face recognition, expression recognition, and pedestrian detection. The experimental results on UC Merced 21-class scene dataset, AR, JAFFE and INRIA pedestrian databases show that our method achieves superior performances over several state-of-the-art methods in terms of recognition rates.

To improve the robustness of image watermarking algorithms against geometric attacks such as cropping, rotation, scaling, shearing, projective mapping, a robust image watermarking algorithm based on DT CWT and Schur decomposition is proposed. Firstly, we use Dual Tree Complex Wavelet Transform (DT CWT) to decompose the U component of the original image (cover image), the decomposed low frequency component is divided into non-overlapping N × N blocks, and Schur decomposition is applied in each block. Secondly, we modify maximum primary diagonal value of the upper triangular matrix of each block to embed the watermark. Finally, the Scale Invariant Feature Transform (SIFT) features of the watermarked image (stego-image) is saved as the SIFT key for watermark extraction. When extracting the watermark, we first use the SIFT features of stego-image and changed stego-image to obtain the geometric torsion, thereby a spatial synchronization can be guaranteed. And we then use the maximum election statistics to improve the quality of the extracted watermark. Experiment results show that the proposed algorithm can resist geometric attacks, and is more robust than the existing watermarking methods.

Denoising of hyperspectral images (HSIs) is an important preprocessing step to enhance the performance of its analysis and interpretation. In reality, a remotely sensed HSI experiences disturbance from different sources and therefore gets affected by multiple noise types. However, most of the existing denoising methods concentrates in removal of a single noise type ignoring their mixed effect. Therefore, a method developed for a particular noise type doesn’t perform satisfactorily for other noise types. To address this limitation, a denoising method is proposed here, that effectively removes multiple frequently encountered noise patterns from HSI including their combinations. The proposed dual branch deep neural network based architecture works on wavelet transformed bands. The first branch of the network uses deep convolutional skip connected layers with residual learning for extracting local and global noise features. The second branch includes layered autoencoder together with subpixel upsampling that performs repeated convolution in each layer to extract prominent noise features from the image. Two hyperspectral datasets are used in the experiment to evaluate the performance of the proposed method for denoising of Gaussian, stripe and mixed noises. Experimental results demonstrate the superior performance of the proposed network compared to other state-of-the-art denoising methods with PSNR 36.74, SSIM 0.97 and overall accuracy 94.03 %.

Underwater images play an irreplaceable role as one of the carriers of underwater information acquisition. Underwater degraded images are usually affected by the color cast, noise, and blurred details, which are difficult to apply to various vision tasks. We propose a multi-scale retinex adaptive grayscale transformation underwater image enhancement method, which includes three parts: color correction, image denoising, and detail enhancement. Firstly, the multi-scale Retinex algorithm is adopted to extract the lighting components. Mean and mean square errors were introduced through linear quantization, and color recovery factors were adopted to adjust the three channels for color correction. Second, by treating the image as an anisotropic thermal field diffusing in all direction,image noise is eliminated and edge details are preserved. Finally, for different underwater degraded images, a simulated annealing optimization algorithm is introduced to perform adaptive gray-scale transformation on the image to enhance image details. The results show that the proposed method can comprehensively solve the problems of color distortion, noise, and low contrast. Compared with the state-of-the-art underwater image enhancement and restoration methods, our method has achieved better visual effects.

This paper presents a two-stage sequential method for noisy grayscale and color images. At first proposed method enhances the accuracy of the noise detection stage by using spatial domain filter rank-order absolute difference trimmed global mean (ROAD-TGM) along with transform domain-based progressive principle component analysis (PPCA) method. Then the performance booster algorithm is used to ensure the proximity of restored value to the original value. Quite often in real-world applications images are corrupted with Salt & Pepper noise, strip lines artifact and Blotches artifact. We observed the proposed method is capable of removing the above-mentioned noises and artifacts with comparatively better accuracy. The proposed method uses the progressive PCA for its dimension reduction ability and local information of image restored by ROAD-TGM to provide enhanced noise detection performance. Before noise removal, a performance booster algorithm eliminates noisy values by using sequential hard thresholding and estimates the tentative original values automatically. Then algorithm decides the suitable value for the restoration of noise pixel by using a structural similarity index (SSIM) to ensure the proximity of the restored image to the original image. The proposed algorithm is tested on a standard set of color and grayscale images to ensure the versatility of the proposed algorithm. The experiment shows that the proposed algorithm achieves high denoising performance for noise and artifact while maintaining the visually important details.

The aim of multifocus image fusion is to fuse two or more partially focused images into one fully focused image. To overcome the problem of a limited depth of field and blurred imaging of objects beyond the depth of field in optical imaging systems, a multifocus image fusion method based on a convolutional elastic network is proposed. Each source image is first decomposed into a base layer and a detail layer using the fast Fourier transform. Then, the convolutional elastic network performs fusion of the detail layers while applying the “choose-max” fusion rule to the base layers. Finally, the fused image is reconstructed by a two-dimensional inverse discrete Fourier transform. To verify the effectiveness of the proposed algorithm, we applied it and seven other popular methods to sets of multifocus images. The experimental results show that the proposed method overcomes the shortcomings of low spatial resolution and ambiguity in multifocus image fusion and achieves better contrast and clarity. In terms of both subjective visual effects and objective indicators, the performance of our method is optimal in comparation with other state-of-the-art fusion methods.

Superpixel segmentation is a popular image preprocessing technology in image processing. Among the various methods used to calculate uniform superpixel, the performance of linear spectral clustering (LSC) is better than the state-of-the-art superpixel segmentation algorithms. However, this method is slow on images and has low accuracy on non-convex images. In order to improve this problem, we propose a subsampled clustering method that can accelerate LSC. Meanwhile, this paper presents an improved distance measurement method based on non-convex image features and Manhattan distance, which can achieve high accuracy on non-convex images. The proposed method is evaluated on the BSDS500 dataset. The experimental results confirmed that this method runs faster than LSC, and at the same time produces almost the same superpixel segmentation accuracy on the images. In addition, the proposed method improves the accuracy of superpixel segmentation on non-convex images.

Memory management of the hyperspectral image sensor is a challenging issue. The existing hyperspectral image compression schemes play a dominant role in minimizing the cost of storage equipment and bandwidth for data transmission for resource constraints onboard hyperspectral image sensors. Traditionally many transform-based set partition hyperspectral image compression algorithms are proposed, but these compression schemes use the data-dependent link list to keep track of the significant or insignificant coefficients or block cube sets, and the size of the lists increases swiftly with the encoding rate. The data-dependent list management and multiple memory read or write operations slow down the compression scheme. Many attempts had been made to address the memory issue through the replacement of the dynamic lists by the static fixed size state tables. The memory required for the state table depends upon the dimension of the hyperspectral image and at the low bit rates, it requires a lot of memory. This paper presents the novel hyperspectral image compression scheme for the hyperspectral image sensor that eliminates the linked list and state table. The obtained experimental results show that the proposed compression scheme outperforms state of the art transform hyperspectral image compression schemes in terms of coding memory and computation complexity while maintaining the coding efficiency. Due to the low complex nature, the proposed scheme saves the operation time and energy for the coding operation. The proposed compression scheme is a suitable candidate for the lossy data transmission and for the low memory hyperspectral sensors.

Image encryption is gaining popularity due to the rapid use of internet technology. Security attacks on the image can occur when images are shared through the internet. E-governance generates important documents with images like identity cards, admit cards, and medical reports. An attacker can steal the image data. An attacker can modify or alter the documents using their own images to create duplicate documents. So, image encryption is a must to prevent the security and privacy of images. In this paper, a novel image encryption method using elliptic curve pseudo-random number is proposed for image data transfer in E-governance. The encryption method creates the cipher image with secp256r1 elliptic curve points and a map table. The map table consists of bit positions as index numbers. In the decryption phase, the receiver decodes the cipher image complementing index numbers in the map table. The map table is also encrypted and decrypted with a chaotic sequence. Our method gives better results than other image encryption methods, according to bits per pixels, entropy, correlation of adjacent pixels, histogram analysis, peak signal to noise ratio, unified average changing intensity, encryption, and decryption time. The experimental results prove the proposed method’s effectiveness and security.

Segmentation tasks in medical images have always been a hot topic in the medical imaging field. Compared with traditional images, medical images have richer semantics, which increases the difficulty of feature learning. This paper proposes a new end-to-end dual-channel integrated cross-layer residual algorithm (TIC-Net) based on deep learning to fully mine the semantic information between features for medical image segmentation. First, in the encoder, we built a dual-channel network of traditional convolution and dilated convolution using multiple structures to learn different semantic information from the image and from feature fusion and residual calculation to achieve feature joint mining. Second, we added two sets of new integrated modules between the decoder and encoder to fully fuse the global and local features of each layer of the image in the encoder. Finally, in the decoder, we use a cross-layer feature residual fusion strategy to obtain more semantic information. Compared with the existing partial segmentation model, the proposed deep learning algorithm model achieves the best results with the Kaggle and MICCAI datasets.

The aim of low-light image enhancement algorithms is to improve the luminance of images. However, existing low-light image enhancement algorithms inevitably cause an enhanced image to be over- or underenhanced and cause color distortion, both of which prevent the enhanced images from obtaining satisfactory visual effects. In this paper, we proposed a simple but effective low-light image enhancement algorithm based on a membership function and gamma correction (MFGC). First, we convert the image from the RGB (red, green, blue) color space to the HSV (hue, saturation, value) color space and design a method to achieve the self-adaptation computation of traditional membership function parameters. Then, we use the results of the membership function as the γ value and adjust coefficient c of the gamma function based on the characteristics of different images with different gray levels. Finally, we design a linear function to avoid underenhancement. The experimental results show that our method not only has lower computational complexity but also greatly improves the brightness of low-light areas and addresses uneven brightness. The images enhanced using the proposed method have better objective and subjective image quality evaluation results than other state-of-the-art methods.

Cancelable Biometrics is a challenging research field in which a cancelable template corresponding to a biometric is generated without affecting the performance. These cancelable templates can be generated using various methods given in literature. However, the qualitative and quantitative performance of these methods is not satisfactory. To address this concern, here we propose a novel cancelable biometric template generation method based on Random Permutation and Chinese Remainder Theorem. In the proposed approach, cancelable biometric template is generated by employing secret image i.e. the original biometric and cover image(s) i.e. any other image(s). Random permutation provides visual protection against the original biometric but preserves the intensity values. Chinese Remainder Theorem is used as a transformation function on the output of random permutation step so as to ensure the intensity values are not revealed to any intruder. The proposed method generates Cancelable Biometric templates in the form of Secret Shares and is suitable for gray as well as color images. To demonstrate the efficacy of the proposed method, extensive experiments have been performed on eight different datasets viz. Carreira-Perpinan ear (CP) (gray and color), University of Tehran IRIS (UTIRIS) (color), Olivetti Research Laboratory face (ORL) (gray), Mathematical Analysis of Images ear (AMI) (color), Aleix Martinez and Robert Benavente face (AR) (color), and Indian Institute of Technology Delhi iris (IITD) (gray and color). Qualitative and quantitative analysis of the generated templates shows that the performance of the proposed method is best when considered simultaneously than several state-of-the-art methods. The main advantages of the proposed method include (i) no glimpses of the original biometric in the cancelable template, (ii) no storage of original biometric during enrolment (iii) the proposed method works for both gray as well as color images and (iv) no image registration is required.

Quantum image processing is the use of quantum computing to store, transmit, and process digital images on quantum computers. This paper introduces two enhanced quantum image representations to store quantum images. The first enhanced quantum representation based on the flexible representation for quantum images (EFRQI) is an amplitude representation that uses the partial negation operator to store the values of the pixels of $$2^n \times 2^n$$ image in the amplitudes of the qubits. The second enhanced quantum representation based on the novel enhanced quantum representation of digital images (ENEQR) is a basis state representation that uses the CNOT gate to store the values of the pixels in a qubit sequence. The proposed methods have better time complexity and quantum cost when compared with related models in the literature.

In the process of image acquisition, the existing image-based real-time video acquisition system is susceptible to noise and distortion due to the influence of attitude, illumination and other conditions, which reduces the quality and stability of the acquired image, and thus makes it difficult to locate the image feature area. Therefore, the feature extraction technology of real-time video capture based on deep convolution neural network is proposed. Cut out high-quality images by locating reference points in feature connection areas, smooth each part of the image by using mean image filter, extract texture features by using convolution, transform, discrete cosine transform and statistical features, and replace random initialization weights with pre-trained models. In the process of model training and recognition, the methods of feature state division, image preprocessing and observation vector calculation are studied. The experimental results on ORL database verify the effectiveness of the image feature extraction method, which can meet the needs of current real-time video capture.

Images captured by cameras are sometimes contaminated either during acquisition or transmission. Therefore, a preprocessing step is required which reduces noise from images. In this paper, a novel and efficient edge preserving universal noise removal algorithm is proposed which exploits both the local and global characteristics of the neighboring non-corrupted pixels. In the proposed algorithm, corrupted pixels are detected by robust outlying ratio (ROR) and replaced with the weighted sum (local characteristics) of the neighboring non-corrupted pixels in 3 × 3 window and these weights are obtained by solving the kernel ridge regression (KRR) which uses the global mean and covariance (global characteristics). Extensive experimental results demonstrate that our algorithm has better noise removal capability in terms of both objective and subjective evaluation as compared to existing denoising algorithms.

Non-rigid multi-modal image registration remains a challenging task due to the complex intensity relation between the two images to be registered. Especially in the presence of spatially varying intensity distortion, the conventional methods are unable to register the two images. Here, the proposed framework includes two steps. First, a structural representation of each image is computed, and then a conventional similarity metric such as mutual information (MI) or sum of squared distance (SSD) is used to register the two structural images. The structural image is obtained by utilizing a modified second-order entropy image and the gradient information. The proposed framework is tested on some data ranging from simulated to real data. Quantitative and qualitative results demonstrate that applying MI similarity metric on the proposed representation is capable of achieving high accuracy results.

Almost all existing image encryption algorithms are only suitable for low-resolution images in the standard image library. When they are used to encrypt high-resolution images, they will inevitably suffer from inefficiency or program crashes. Inspired by the super-pixel concept in image processing, in this paper, a new image encryption algorithm using PRSP (Pseudo Random Super Pixel) strategy and KBCA (Key-based Cellular Automata) is proposed. Both scrambling and diffusion are implemented in parallel to improve the efficiency of encryption algorithm when encrypting ultra-high resolution images. Besides, random diffusion path is designed to ensure the security of encryption algorithm. Simulations are carried out using CUDA (Compute Unified Device Architecture) platform, and simulation results show the high efficiency and security of our algorithm.

This paper introduces a deep learning-based Steganography method for hiding secret information within the cover image. For this, we use a convolutional neural network (CNN) with Deep Supervision based edge detector, which can retain more edge pixels over conventional edge detection algorithms. Initially, the cover image is pre-processed by masking the last 5-bits of each pixel. The said edge detector model is then applied to obtain a gray-scale edge map. To get the prominent edge information, the gray-scale edge map is converted into a binary version using both global and adaptive binarization schemes. The purpose of using different binarization techniques is to prove the less sensitive nature of the edge detection method to the thresholding approaches. Our rule for embedding secret bits within the cover image is as follows: more bits into the edge pixels while fewer bits into the non-edge pixels. Experimental outcomes on various standard images confirm that compared to state-of-the-art methods, the proposed method achieves a higher payload.

International audience; This paper presents a novel method for image manipulation iden- tification of natural images in JPEG format. The image forgery detection technique is based on a signal-dependent noise model that is relevant to de- scribe a natural image acquired by a digital camera. This parametric model is characterized by two fingerprints which are used to falsification identification. The problem is cast in the framework of the hypothesis testing theory. For practical use, the Generalized Likelihood Ratio Tests (GLRT) are presented and their performance is theoretically established. There are different types of image forgery which have been considered in this paper for example re- sampling, Gaussian filtering and median filtering. Experiments with real and simulated images highlight the relevance of the proposed approach.

This article introduces a robust image watermarking primarily founded on DWT (discrete wavelet transform), BEMD (bi-dimensional empirical mode decomposition), DCT (discrete cosine transform), PSO (particle swarm optimization), and SVD (singular value decomposition). During the process of embedding, 2nd level DWT is used to decompose the cover image into sub-bands. DWT is also used for the decomposition of images for watermarking. Further, BEMD decomposition runs to implement on the selected band of DWT. For optimization, PSO is used for complex and multidimensional searches. Further DCT as well as SVD applied to the selected band. The embedding and scaling factors are embedded with the help of a security key. Further, this method is followed by using the inverse of ISVD, IDCT, IDWT, and IBEMD. The Watermark image is extracted by the extraction process. Experimental results show that the suggested technique is robust compared to several geometrical and non-geometrical attacks. Therefore, this proposed technique provides better visible quality and robustness against numerous attacks like salt and pepper, Gaussian filter, rotation, median filter, speckle, gamma correction, scaling, and shearing for gray scale images and provides the watermarked image with good quality.

Edge detection has many applications in engineering and medical field. Edge detection in color images is getting the attention of the researchers due to the reason that color images have more information as compared to gray scale images. Different differential methods have been proposed in the literature for edge detection. Some of them required smoothing due to high sensitivity of differential methods towards noise. In the present manuscript, fractional order differentiation operator is defined to find out the gradient of the image which is further used for edge detection. Considering the input image as a reconstructed image, optimal threshold selection method is defined which is based on an error in image reconstruction assuming thatthe input image isa reconstructed image. Fractional edge detection improves the thin edge detection. It has more details of edges as compared to integer order based fractional differentiation. Fractional differentiation based edge detector does not require additional smoothing. Optimal threshold selection based on an error in a reconstructed image enhances the texture information as compared to normal threshold selection.

This paper presents a new two-step image denoising method termed multidirectional gradient domain image denoising (MGDID). In each step, unlike previous gradient domain designs, the multidirectional gradient domain information is used to represent the noise component so that the more directional image features are extracted. The Gaussian pre-filter is carried out in the square gradient coefficients. The nonlinear remedied factor is adopted to modify the denoising amount. The whole denoising process originates from classical nonlocal means (NLM) and nonlinear diffusion. MGDID takes full advantage of ability of NLM to better process the image with the rich repetitive features and the denoising scheme of relatively simplicity and efficiency of nonlinear diffusion. Experimental results show MGDID is superior to the related gradient domain methods and NLM methods in peak signal-to-noise ratio (PSNR), mean structural similarity (MSSIM) and visual performance. For example, for Barbara image with the rich repetitive texture feature, MGDID outperforms classical NLM from 0.33 dB to 1.66 dB in PSNR. Usually, classical NLM wins the local adaptive layered Wiener filer (a state-of-the-art gradient domain method) more than 0.44 dB for Barbara. In addition, MGDID is also very efficient compared to the related methods.

Image noises are usually generated in the processes of collection, transmission, and storage of images, while some noises, named internal noises, come from the image itself. The noises decrease image visual effect and quality. Thus, it is very important to remove the noises from the images. In this paper, we propose a one-dimensional surface blur algorithm based on wavelet transform and bilateral filtering for image internal noise elimination and detail preservation. In our algorithm, we first transform the two-dimensional image into one-dimensional signal vectors by merging the pixels in each row and column of the image. Then, we decompose each of the vectors into two parts: the low-frequency and high-frequency components with a discrete wavelet transform. We further perform the bilateral filtering and a local variance-based thresholding method on the two components to smooth and denoise signals, respectively. Finally, we evaluate our algorithm’s performance in a group of face images. The experimental results show that our algorithm achieved better performance on image denoising and detail preservation than a set of traditional smoothing methods and the state-of-the-art. Our algorithm is a simple, effective, and easy-to-implement method, and it is suitable for image smoothing to improve the image’s visual effect and quality.

Due to the serious attenuation and scattering effects of light underwater, actual underwater images have problems such as low contrast and color distortion. This paper proposes an underwater image enhancement algorithm. First, we apply the guided filtering to the algorithm improvement to get the improved image. Then, the image is converted from RGB color space to HSI space, and the three components of hue, saturation, and intensity are separated. Then use adaptive feedback adjustment to achieve the stretching of saturation and linear enhancement of intensity. Then the image is converted from the HSI color space back to the RGB color space to obtain an enhanced image. Finally, the improved image and the enhanced image are merged at the pixel level. After experimental analysis and comparison, the time required for guided filtering to process images can be reduced by 65%, the structural similarity can reach more than 90%, and the peak signal-to-noise ratio and information entropy have been greatly improved. From a visual point of view, the color saturation, color richness, local contrast and clarity of the image have all been significantly improved.

Nowadays, the numerical segmentation is an important step in the processing and interpretation of medical images. The segmentation consists in extracting, from the image, one or more objects forming the regions of interest. Image thresholding is one of the simplest and effective techniques of image segmentation. In this work, we propose and compare multilevel segmentation approaches based on classical and quantum techniques. The Classical Renyi (CR) and the Quantum Renyi (QR) entropies are used to quantify the information contained in the image. Within the quantum framework, the digital image is expressed as a quantum system by means of the Flexible Representation of Quantum Images (FRQI). Generally, the multilevel thresholding formulation leads to a complex optimization problem. The Classical Genetic Algorithm (CGA) and the Quantum Genetic Algorithm (QGA) are employed to efficiently determine the optimal thresholding values by maximizing the entropy-based fitness functions. The segmentation performances of the proposed methods are assessed and compared using some prevailing criteria. The achieved results on a sample of medical images demonstrated that the QGA-QR method outperforms significantly the other combinations for this thresholding exercise.

Infrared and visible image matching has many applications in remote sensing, computer vision, military fields, etc. The differences in the many characteristics of infrared images and visible images make a robust feature description vital but difficult. Texture orientation information retains the general properties in refrared and visible images, and multi-scale, and multi-oriented Gabor filters can accurately reveal the texture orientation information. This paper presents a feature descriptor by capturing the phase information between neighboring pixels with Log-Gabor filters. Firstly, the original matching image is enhanced via histogram equalization to emphasize the regions of interest, and the gradient magnitude for each pixel is computed to extract the image profile, which advances the performance of the algorithm significantly. Secondly, multi-scale and multi-oriented Log–Gabor filters are utilized to obtain the angle information for different scales and phases in the neighboring region of each pixel, and the angle information is indexed by computing the maximum of energy, including the magnitude, real part, and imaginary part to generate the marked image in which the histograms of the subregion of the detected keypoints are employed to generate the feature descriptors. Finally, we advocate five evaluation measures for testing the performance of the algorithm. The proposed approach is evaluated with four data sets composed of images obtained in visible light and infrared spectra, and its performance is compared with the performance of the state-of-the-art algorithms: Scale-invariant feature transform(SIFT), Speeded up robust features(SURF), Oriented fast and rotated BRIEF(ORB), the edge-oriented histogram descriptor (EHD), the phase congruency edge-oriented histogram discriptor (PCEHD), and the Log–Gabor histogram descriptor (LGHD). The experimental results indicate that the performance of the proposed approach is higher than that of other state-of-the-art algorithms.

In this article, the concept of a complete 3-Partite graph is used to substitute the pixels of a color image to achieve encryption. Chen’s chaotic system is deployed to generate vertices of a tri-partite directed graph and then paths are established between these vertices. These paths will serve to perform the substitution of individual pixels. The seeds for the chaotic maps are manipulated by the hash value of the plain image to hinder chosen plaintext attack. The inter and intra correlation of pixels of a three color image in one-dimensional vector is distorted using the non-linear system of equations. The different chaotic maps having divergent features are employed to add up complexity in encryption technique. The extensive experiments are applied on different kinds of images and results are compiled. The results show that the proposed system has a strong robustness against different attacks and prove to be an excellent candidate to encrypt colored digital images. It has high scores for the Number of Pixel Change Rate (NPCR), Unified Average Cipher Intensity (UACI), entropy values, low values for Mean Absolute Error (MAE), Chi-Square and variance of histograms for encrypted images. The palpable lead of proposed system has resilience against transmission impairments such as Gaussian, Salt & Pepper and clipping.

Nowadays, dictionary learning has become an important tool in many classification tasks, especially for images. The tailor-made atoms in a dictionary are trained for the reconstruction of the test sample. In the classification, atoms are associated with different classes from several subspaces such that the test sample is labeled according to the distances of each subspace. However, it is hard to fix the number of atoms to obtain the optimal result for each scenario since the optimal subspaces required are different. To improve the classification performance as well as the robustness, we proposed subspace-level dictionary fusion (SLDF) to construct a dictionary-based classifier. A full-size dictionary and a locality-constrained dictionary are constructed in parallel. Then, the reconstruction coefficients of the two dictionaries are obtained, which leads to a pair of distances between the test sample and the subspaces. Finally, a decision is made according to the pair-wise fusion of the distances. The experimental results on multimedia datasets from distinct categories such as image, text, and audio show that the proposed method outperforms other state-of-the-art dictionary-based classification methods with accuracies of 99.74% (image), 83.96% (Text), and 87.07% (Audio).

In this work, we develop a new approach for learning a deep neural network for image classification with noisy labels using ensemble diversified learning. We first partition the training set into multiple subsets with diversified image characteristics. For each subset, we train a separate deep neural network image classifier. These networks are then used to encode the input image into different feature vectors, providing diversified observations of the input image. The encoded features are then fused together and further analyzed by a decision network to produce the final classification output. We study image classification on noisy labels with and without the access to clean samples. Our extensive experimental results on the CIFAR-10 and MNIST datasets demonstrate that our proposed method outperforms existing methods by a large margin.

Image segmentation is a method of subdividing an image into numerous meaningful regions or objects, which shows the image more informative for further analysis. Thresholding based methods are extensively used for image segmentation due to its easy implementation and low computational cost. However, histogram-based thresholding techniques are unable to deliberate three-dimensional contextual information of the image for optimum thresholds. In this paper, energy-curve is coupled with 3D Otsu function. Furthermore, in order to increase the quality of the processed image, a simple and effectual approach is proposed by using the concept of fusion, grounded on local contrast. The presentation of 3D Otsu algorithm is described to be poor when dealt with between-class variances over the aid of 3D histogram. To alleviate this limitation, the perception of the energy curve has been used to derive pixel intensity values and spatial information. Energy curve can help to recover the excellence of the thresholded image as it computes not only the value of the pixel but also its vicinity. The proposed energy based 3D Otsu with fusion (3D-Otsu Energy Fusion) method uses exhaustive search process to determine the optimal threshold values. The proposed technique produces better-processed results as compared to rest methods.

Weber’s law reveals the relationship between human perception and perceptual stimuli. Inspired by the theory, this paper designs a multi-level convolution correlation feature statistic method for image retrieval. Firstly, the difference between a central pixel and its neighbors is described by Weber’s law through computing the differential excitation of image. Then, a multi-level saliency map is obtained by binary transformation and convolution operation. Thirdly, to exploit spatial correlation information of the image, a pixels pair-wise correlation and hierarchy statistic model is constructed. Finally, all intermediate features are concatenated into one histogram, which includes salient color and texture features. Extensive experiments demonstrate the proposed method of this paper has excellent performance.

This paper presents a two-step fuzzy filter to remove impulse noise from a color image sequence in RGB color space. The primary filter recognizes the pixels corrupted by impulse noise. It also computes the extent of noise and afterward rectifies them. The output of the first filter acts as an input to the secondary filter and further refines the outcome to give the final output. Excellent alignment is seen between noise removal and structure conservation of an image due to the classifying nature of the algorithm. To minimize blurring, noisy pixels are exclusively rectified and noise-free pixels remain intact. The proposed filter is a 3-D Spatio-temporal filter that considers spatial, temporal as well as color information. A pixel of one color component is compared to its neighboring pixels within the same frame and with the corresponding pixels in neighboring frames. It is likewise compared with the pixels of the other two color components. Peak Signal to Noise Ratio (PSNR) and structural similarity index (SSIM), Miss Detection (MD), and False Alarm (FA) rates are utilized as a performance metric. The experimental result of several color image sequences demonstrates the efficacy of the proposed fuzzy filter both qualitatively and quantitatively.

Aiming at the shortcomings of traditional Retinex image enhancement algorithms, such as poor texture detail retention, halo, over-enhancement and hue mutation, a low-illuminance color image enhancement algorithm based on Gabor filter and Retinex theory is proposed. The algorithm extracts the luminance I component from the HSI color space of the original image, and then performs MSRCR (Retinex algorithm for color restoration) enhancement on the luminance I component to obtain an enhanced luminance I component and color reproduction image. On the other hand, the original image is enhanced by a SSR (Single Scale Retinex Algorithm) based on the Gabor filter in the RGB color space to obtain an enhanced image with better texture and edge details. Then, the two images enhanced in two different ways are weighted and merged to obtain the final enhanced image. This algorithm is compared with the SSR algorithm based on Gamma correction, the MSR (multi-scale Retinex algorithm) based on bilateral filtering and the improved MSRCR algorithm. Taking mean square error, information entropy, and average gradient as evaluation indicators, the experimental results show that the image information processed by this algorithm is rich in color, rich in color, and color is closer to the original image, effectively reducing the occurrence of halo and excessive enhancement. This algorithm has certain reference significance for the enhancement of low-light color images.

Some image encryption algorithms are difficult to resist the chosen-plaintext attack against special images, in order to solve this problem and improve the security of the algorithm, this paper proposes a novel image encryption scheme based on the chaotic system, random number embedding and DNA-level self-adaptive permutation and diffusion. The architecture of preprocessing, permutation and diffusion is adopted. Firstly, an image preprocessing based on random number embedding (IPRNE) is presented, specifically, embed random numbers into the plain image, and then perform partition XOR operation on random numbers and their surrounding pixels to preprocess plain image. The random numbers are generated by a 4D memristive hyperchaotic system, and their embedding positions are controlled by the pixel sums of plain images. Secondly, the obtained image is encoded into a DNA matrix by use of a DNA encoding rule, and then a DNA-level self-adaptive permutation and diffusion processes are successively performed on it. Further, after decoding the diffused matrix, the cipher image is obtained. Besides, the feature information of DNA sequences of plain image is applied for disturbing the permutation and diffusion phases, which may be extracted automatically in the decryption process, and thus additional transmission and storage are avoided. Moreover, the plain image information and hyperchaotic system are integrated to design the DNA encoding /decoding rule for the plain image and mask matrix, and this can enhance the ability of the algorithm to resist chosen-plaintext attack. Experimental results and security analyses demonstrate that the proposed encryption is secure and effective, and it can be applied for image secure communication.

The diffusion-based graph has been widely used in saliency detection. Most of the existing methods treat the image boundary patches as background seeds, which may result in the imprecise saliency map if the salient region touches the image boundaries. In this paper, we propose a salient object detection approach via cross diffusion-based compactness on multiple graphs. Firstly, we extract multi-view image features including low-level image features, mid-level image cues (low-level saliency priors), and background-based saliency map. Then, we compute the respective similarity matrix to construct the corresponding graph, and a cross-diffusion algorithm is presented that diffuses each similarity matrix on other’s graphs rather than at itself graph, helping to the compactness-based saliency maps. Additionally, for well propagating the saliency values, we model a propagation mechanism based on cellular automata, by linearly incorporating low-level image features and mid-level image cues together to generate a superior impact factor matrix. Extensive experiment results demonstrate that the proposed method achieves better saliency detection performance against the unsupervised state-of-the-art methods on three public datasets.

At present, most denoising algorithms cannot determine whether a pixel is noise, but these use the same rules to process all pixels. Most denoising methods will filter out the original image information when they deal with images with more details or little difference between the subject and the background. In order to improve the above shortcomings, a two-stage image denoising algorithm of noise localization in this paper is proposed.Firstly, the thresholds $$ {T}_1^{\prime } $$ and $$ {T}_2^{\prime } $$ are extracted according to the image gray value distribution. Image edge information is removed and saved by edge extraction, this gets an edgeless greyscale image. Secondly, singular value decomposes the edgeless image to obtain the singular value matrix, the percentage threshold η is used to reduce the singular value matrix.The coarse noise filtering is performed by the inverse matrix decomposition. Again, the adaptive thresholds T1 and T2 are calculated with the histogram, the image is divided into “Dark Area”, “Gray Area” and “Light Area”. Then, a superpixel-like algorithm is introduced to determine and remove the noise accurately in three regions. Finally, the image edges are combined with the denoised image. By analyzing the denoising image and comparing the peak signal-to-noise ratio (PSNR) and time of the result in many images, it is verified that the proposed algorithm has a better denoising effect than many other denoising algorithms.

In order to improve the key space of color image encryption algorithm, the sensitivity to the contents of plain images, the robustness against various types of known attacks, and to achieve the tamper location analysis, a novel color image encryption algorithm based on image hashing, six-dimensional (6D) hyperchaotic and dynamic DNA coding is proposed. Firstly, the color image is pre-processed and the hash sequence is extracted by image hashing algorithm which is used as the initial value and control parameter of chaotic system. Secondly, three color channels of the color image RGB are synthesized into a two-dimensional matrix and the pixels replacement are performed by using the improved two-dimensional chaotic map. Finally, a 6D hyperchaotic system is used to generate random sequences for DNA dynamic coding and arithmetic operations of color images, the encrypted image is obtained. The experimental results show that, compared with the existing methods, the proposed algorithm has a large enough key space, better plain-image sensitivity, better statistical and differential characteristics, as well as can resistant various forms of attacks such as noise and cropping, and the tampering image can be tamper-located analyzed, which has good security and strong robustness.

Machine learning based saliency detection methods have achieved better performance than traditional methods. Here, we propose a machine learning based method that utilizes Convex Hull and Extreme Learning Machine (ELM) for detecting salient object(s) in an image. The novelty of the proposed method lies in the generation of training set without using human annotations. Initially, an input image is segmented using SLIC algorithm at different scales to produce multiscale segmented images. This is followed by estimating two different saliency priors viz. (a) Convex Hull center prior and (b) contrast prior for each segmented image. These priors exploit foreground center and spatially weighted contrast respectively. Both of these estimated priors help in computing initial saliency of each segment across all scales. For each scale, the initial saliency map along with the Convex Hull based label map is employed on a segmented image to determine the positive (salient) and negative (background) training set. Distinctive features for each segment belonging to the training set are extracted and then passes to the Extreme Learning Machine for learning the ELM model. Afterwards, multiscale saliency maps of an image are found by applying the learned ELM model on distinctive features extracted from each segment across multiple scales. These multiscale saliency maps are linearly combined to obtain the final saliency map. The effectiveness of the proposed method is supported through extensive experimental results performed on six publicly available datasets viz. MSRA10K, DUT-OMRON, ECSSD, PASCAL-S, SED2, and THUR15K. The performance of the proposed method was compared with 11 state-of-the-art methods in terms of Precision, Recall, F-Measure, Receiver Operating Characteristics (ROC), and Area under the curve (AUC). The proposed method outperforms or comparable with compared methods in terms of all the performance measures.

In this paper, a new scrambling algorithm is proposed to reduce the correlation between the image pixels in three directions. In this paper, this algorithm is called ladder scrambling, and the image is scrambled multiple times in the form of diagonal lines. In terms of diffusion, DNA coding is used, and the secondary processing of the image is performed through DNA coding to change the pixel value of the image. The final encrypted image is affected by three conditions: the number of scrambling algorithms, the selection of DNA coding rules, and the key of the chaotic system. Theoretical analysis and experimental results show that the algorithm is safe and effective.

Multi-spectralimages suffers from out-of-focusblur due to well focused camera at the reference imaging channel. A framework for out-of-focus images de-blurring using texture extraction and modified Fourier transform is proposed. The texture is extracted from the blurred image using region covariance. Fourier transform is modified by modification of guided image (as prior) using L0 gradient projection followed by detail amplification which is used as an input for Fourier transform. For further enhancement, the multi-spectral de-blurred images are smooth, and ringing artifacts are minimized by combining with the textures extracted to ensure maximum edge preservation. Comparison of proposed and existing schemes on different multi-spectral images explains the advantage of the proposed scheme for de-blurring in terms of edge preservation, noise and artifacts removal.

In this work, we propose a novel multiscale transform decomposition model for multi-focus image fusion to get a better fused performance. The motivation of the proposed fusion framework is to make full use of the decomposition characteristics of multiscale transform. The nonsubsampled contourlet transform (NSCT) is firstly used to decompose the source multi-focus images into low-frequency (LF) and several high-frequency (HF) bands to separate out the two basic characteristics of source images, i.e., principal information and edge details. The common “average” and “max-absolute” fusion rules are performed on low- and high-frequency components, respectively, and a basic fusion image is generated. Then the difference images between the basic fused image and the source images are calculated, and the energy of the gradient (EOG) of difference images are utilized to refine the basic fused image by integrating average filter and median filter. Visual and quantitative using fusion metrics like VIFF, QS, MI, QAB/F, SD, QPC and running time comparisons to state-of-the-art algorithms demonstrate the out-performance of the proposed fusion technique.

The extraction of feature vectors of the images is the most important step in image recognition, because it allows a good description of the forms to be recognized. This step is an operation that makes it possible to convert an image into a vector of real or complex values that can serve as a signature for this image. For an accurate recognition system, the used feature vector must be invariant to the three image transformations (translation, rotation and scale), which means that, the descriptor vectors of the image and the transformed image by translation, rotation or scale must be equal. Given the importance of moments and their invariants in pattern recognition and imaging, based on Newton’s binomial and trinomial formulas and the normalized central moments, we construct in this paper four new series of moments for 2D and 3D image recognition, which are invariant to translation, scaling and rotation (TSR): the first is a set of non-orthogonal moment invariants for 2D images (2dMIs), the second is a set of orthogonal moment invariants for 2D images (2dOMIs), the third is a set of non-orthogonal moment invariants for 3D images (3dMIs) and the fourth is a set of orthogonal moment invariants for 3D images (3dOMIs). Using the proposed invariant moments (2dMIs), (2dOMIs), (3dMIs) and (3dOMIs), we construct two types of image descriptor vectors for 2D images and two types for 3D images, which are invariant to translation, rotation and scale. A series of experiments is performed to validate this new set of invariant moments and compare its performance with the existing invariants moments. The obtained results ensure the superiority of the proposed moments over all existing moments in image recognition. Experiments on processing time show that the proposed method is faster than the existing orthogonal invariant moments.

Some compressed images using Vector Quantization algorithm suffers from blocking artifacts which degrades the visual appeal of the image. Present study proposes a hybrid vector quantization method applicable on de-correlated color model. As luminance channel carries image information and loss of image information results in degradation of the visual appeal of an image, so aim of this study is focused on retaining more image information during compression process. For luminance channel compression, a new four level quantization based compression method is developed. Luminance channel is partitioned into smaller blocks. Then for each block, four level quantization is applied which local to the current block only. This results many level luminance value effectively for the whole image. It helps to retain better information. Chrominance channels are compressed using conventional Vector Quantization. This hybrid compression method improves visual quality of the decompressed image reasonably compared to VQ. The proposed method is applied on many standard images found in literature and images of UCIDv.2 color image database. Results are analyzed in terms of Peak Signal to Noise Ratio, Structure Similarity Index and space requirement reduction for compressed image using the method. Experimental results show that proposed method retains better quality of image in terms of PSNR and SSIM than Vector Quantization and Modified Vector Quantization. This method reduces storage space requirement for the compressed images in the range of 84% to 89%.

Underwater environment is complex and random. The images obtained from underwater by sonar always have uneven background gray distribution and fuzzy details of boundary. Hence the low-quality sonar images need to be enhanced before analysis. This paper presents a sonar image enhancement algorithm based on the approximation spaces of random sets. First of all, the knowledge representation of the underwater image is constructed by the approximation spaces of random sets. According to the background knowledge, the image is divided by the upper and lower approximation space of the random set. Then the optimal partition is obtained according to the approximate equivalence relation of the upper and lower approximation. Based on the optimal partition, an improved dark channel theory is presented to enhance each region of the image. After that, sonar images with different backgrounds are used to test the proposed method. The experimental results show that the gray distribution of the sonar image enhanced by this algorithm is more uniform and the boundary details are clearer. The proposed algorithm has the advantage of solving the optimal division for the set of pixels with approximate grayscale. Moreover, the proposed algorithm can get better image enhancement effect in the premise of maintaining the texture of the sonar images.

Image segmentation is a technique of subdividing an image into numerous sections that converts an image into more expressive form that is easier to analyze. Histogram is one of the most widely used techniques for segmenting a digital image due to its simplicity. However, this method often leads to unsatisfactory segmentation performance because of abnormalities on gray level histogram. In this paper, we propose a technique for segmenting a digital image through multilevel iterative variational mode decomposition (VMD) using Renyi entropy. The VMD is employed first in order to decompose the gray-level histogram into corresponding sub-modes for analysis and attributes extraction. Splitting gray level histogram into various modes results in removal the unfavorable effects. Then, Renyi entropy is applied in order to find best threshold value for image segmentation. The feature set has been formulated by applying non-linear Renyi entropy on each of the modes extracted using VMD. The proposed technique has been tested on standard images and the experimental outcomes indicate that it can produce judicious segmentation outcomes compared to other techniques.

The Zernike and pseudo-Zernike moments (ZMs/PZMs) have been found useful for a variety of applications requiring feature extraction due to their favorable properties, such as low information redundancy, rotational invariance, higher noise resilience and extendibility to the color space. In this paper, we propose an approach for color face recognition based on Zernike/pseudo-Zernike quaternion moment vector (QMV) features and a novel normalized-discriminant hybrid color space. The proposed XnSBr color space is composed by taking Xn from the normalized-XYZ, S from HSV and Br from the discriminant RGB-r color spaces to capture the features efficiently. In addition, we propose the use of quaternion vector distance (QVD) similarity measure for the QMV features in order to enhance the recognition accuracy. The exhaustive comparative performance analyses with the state-of-the-art approaches in the different color spaces demonstrate the superiority of the proposed approach in terms of accuracy and speed.