Search

Digital data are rising fast as Internet technology advances through many sources, such as smart phones, social networking sites, IoT, and other communication channels. Therefore, successfully storing, searching, and retrieving desired images from such large-scale databases are critical. Low-dimensional feature descriptors play an essential role in speeding up the retrieval process in such a large-scale dataset. A feature extraction approach based on the integration of color and texture contents has been proposed in the proposed system for the construction of a low-dimensional feature descriptor. In which color contents are quantified from a preprocessed quantized HSV color image and texture contents are retrieved from a Sobel edge detection-based preprocessed V-plane of HSV color image using a block level DCT (discrete cosine transformation) and gray level co-occurrence matrix. On a benchmark image dataset, the suggested image retrieval scheme is validated. The experimental outcomes were compared to ten cutting-edge image retrieval algorithms, which outperformed in the vast majority of cases.

In recent years, the COVID-19 and its variant are more dangerous forpeople with some health complexity, such as breast cancer, diabetes, andheart disease. Diagnosis at the early stage of these diseases may savepeople from deadly diseases and infections. However, often these diseasesare diagnosed very late. Classification algorithms are in use forthe diagnosis of several diseases. Classification result significance is amajor issue in many real-time domains which deal with class imbalancedata. Class imbalance data is one of the primary challenges for theperformance of classification models. Most sensitive and essential realtimedomain data are class imbalances, resulting in poor classificationperformance. Due to the dominancy of the major class, the classificationof the minor class may show high accuracy. Misclassification cannot be accepted in medical cases, which can cost human life as well.The paper presents a hybrid model named PNM to address the classimbalance issue and thus improve the performance of the classifiers.This proposed sampling model integrates Principal Component Analysis(PCA) and the Near-Miss sampling method. Experimental resultsshow the efficiency of the PNM model for the class imbalance data.

Humans can easily spot people, objects, scenes, and visual details in photographs and videos. The main goal is to make a computer capable of what humans can do with learning a level of understanding of actual image/video content. Object recognition is one of the computer vision techniques for identifying objects in images or videos, and it is a key learning point. The proposed research addresses unsupervised few-shot object recognition learning, where all the supervised training images are prepossessed. In this learning, the classes are shared partially with labeled support images for few-shot recognition in testing. This work uses a new SVM-like deep architecture for unsupervised learning of an image representation, where it will encode latent object parts that generalize well to partially shared classes in the presented few-shot recognition task. The proposed model recognition task integrates self-supervision training decomposition in few-shot learning. This paper is based on two stages; rstly, to handle the discussed issues, there is a need for an intelligent mechanism to observe, detect, and predict with speci c control parameters. Secondly, this study has used at observation level singular value decomposition (SVD) and zero-shot learning for detection and prediction tasks. The large-scale CIFAR-10 image dataset has been used in the experimental results to validatethe proposed method. Several Experiments have been performed on the given dataset, and the eciency of the proposed method has been remarkable over some state-of-the-art classi cation algorithms.

Animal biometrics based cattle identification schemes have been widely explored to overcome the challenges of classical methods. But the existing schemes struggle with accuracy rate and processing time which are the key issues. In this paper, we propose a muzzle pattern based cattle identification scheme to achieve higher accuracy rate using generative adversarial networks (GAN) for image enhancement and coupled SIFT with RANSAC for robust feature extraction and matching. We have focused on muzzle image enhancement to overcome the artifacts of image acquisition and thus facilitate efficient feature extraction and matching. SIFT is used for feature extraction and matching during enrolment and identification process. RANSAC further optimizes the matching. Experimental results show that the proposed scheme achieves identification accuracy rate is 98.86% with optimum processing time. Comparison with existing state-of-art schemes shows the superiority of the proposed scheme.

Due to the tremendous increase in the digital image data, the efficient and effective image content-based search scheme for retrieving desired images from a large image repository is highly required. The biggest challenge in image retrieval scheme is to retrieve the desired multimedia images from the digital image repository with minimum time. Extracting significant image features with low dimensional feature descriptor play a significant role in improving retrieval outcomes. In the presented paper, an image retrieval scheme is proposed using fused low dimensional feature descriptor which is obtained by fusion of probability histogram-based HSV color moments and multiresolution based shape moments. The color moments and shape moments are extracted from the Laplacian filter based preprocessed image. The suggested scheme is implemented on a standard Corel-1K image dataset and the retrieval accuracy is measured using precision, recall, and F-score metrics. The experimental outcomes are also validated and compared with some existing state of the art image retrieval schemes and it outperforms over the existing ones.

In content based image retrieval(CBIR), the searching and retrieving of similar kinds of digital images from an image database are realized on the visual features of a given query image. The efficiency and accuracy of any CBIR scheme depends on the extracted significant visual features of the digital images. This paper considered a CBIR scheme based on the proficient combination of extracted color and texture visual features. The visual features are extracted from the enhanced HSV color image after enhancing the RGB color image using Laplacian filter. In the presented work, the color feature is extracted from the quantized histograms of Hue (H) and Saturation (S) components while texture feature is extracted from computed gray level co- occurrence matrices (GLCMs) of each sub image of discrete wavelet transform (DWT) of Value (V) component of HSV color image. The extracted color and texture visual features are combined together after normalizing them individually. The proposed CBIR scheme is evaluated on standard Corel image database and observed that the combined feature vector produces the satisfactory results in terms of performance evaluation metrics i.e. precision, recall and F-score. The experimental results are also showed that the proposed CBIR scheme outperforms as compare to the some other existing schemes.

Content based image retrieval (CBIR) is the process of searching similar images from an image database based on the visual contents of the input query image. In CBIR, color is considered as one of the prominent features of the image data, so in this paper, the authors have presented a CBIR technique using color based feature. Since a color image, consists of three basic color components, i.e. red, green and blue, so in this work, we have given the same importance on all three color components during image retrieving process. In the presented CBIR technique, initially we have constructed three probability histograms for each color component and subsequently the histograms are divided into several numbers of significant bins and from each bin, we have computed several statistical values like standard deviation, skewness and kurtosis. The computed statistical values are used as extracted features of the image data. The processing cost of the presented CBIR technique is significantly low. The technique has been tested on standard image databases and satisfactory results have been achieved.

In this paper, a content-based image retrieval scheme using the singular value decomposition (SVD) is proposed where the feature vector was estimated from the selected significant components of a singular value decomposed image. The Eigen values of the decomposed image are divided into several numbers of groups and from each group we have constructed several Eigen images and subsequently, statistical values like mean, standard deviation and entropy are computed from those Eigen images. The constructed Eigen images are suitable to analyse the original image data in various image planes. This approach is applied to each colour components for formation of colour-based final feature vector. This approach is appropriate to reduce the overall processing cost in image retrieving process due to the consideration of significant image feature in SVD domain. The scheme is tested on a standard Corel image database and satisfactory results are achieved.