Search

This study analyses the use of a pixel classification model to segment amniotic fluid areas on ultrasound (US) images characterized by noise, blurry edge, artifacts, and low contrast. In contrast with the previous methods, this study constrains a training set of pixels based on neighbourhood information with the rectangle window sampling method used to determine the characteristics of each pixel in its environment specifically. The feature extraction is no longer based on the global characteristics of the object rather by taking the value of each pixel in the object area using the sampling window. This research also combines the local first-order statistical methods and gray level information in the window to obtain the characteristics of each pixel. Furthermore, Random Forest and Decision Tree (C.45) were used to classify each pixel into four classes, namely amniotic fluid, placenta, uterus, and fetal body. The classification performance testing of pixel sampling data showed that the Random forest with 5 × 7 window sizes achieved the highest performance at 99.5% accuracy, precision, and recall, respectively. Furthermore, the proposed model was evaluated using 50 new test US images to segment the amniotic fluid area. According to experimental result, proposed models can produce better segmentation area with an increase in the IoU value by 18.3% or a Jaccard coefficient value rate of 0.183 in the range of 0-1 with the previous state of the art method. Furthermore, the proposed model reduces the error rate and improves accuracy by 6.61% and 84.77%, respectively.

Hyperspectral band selection (BS) aims to select a subset of bands from the original image cube for subsequent tasks, such as pixel classification. In this paper, we propose a novel unsupervised BS...

Semantic image segmentation treats the issues involved in the object recognition and image segmentation as a combined task. The chief notion of semantic segmentation is to partition the image into visually uniform regions and to discriminate the class of the partitioned regions. Pixel classification is done over the segmented regions by assigning semantic labels. In general, inference frameworks are fed with the combination of low‐level features and high‐level contextual cues to segment an image. Since these combinations are rarely object consistent, result with minimum classification accuracy because of choosing non‐influencing features and cues to track specific objects. To overcome this problem, a nature‐inspired meta‐heuristic optimization algorithm called Seed Picking Crossover Optimization (SPCO) is proposed to optimize i.e. train the CRF (Conditional Random Field) for choosing relevant feature to segment the object with high accuracy. To meritoriously recognize the objects, a semi‐segmentation process is initially performed using Simple Linear Iterative Clustering (SLIC) algorithm. For pixel transformation and pixel association, Dirichlet process mixture model and CRF are employed. Optimized CRFs are used where the parametric optimization is done using the proposed SPCO algorithm. The proposed work results with 84% on classification accuracy and the performance evaluations are done using MSRC‐21 dataset.

The hyperspectral image such as AVIRIS-NG provides a lot of spectral information and fine resolution that identify and discriminate similar objects based on their spectral reflectance. This study r...

In order to solve the problem of difficult pre-processing of crop video image shadows, a probable learning pixel classification method is proposed to study its processing technology. The algorithm effectively detects the shadow area by performing intelligent video collaborative detection on the shaded parts of the crop video sequence. Firstly, the cloud collaborative detection algorithm that can be widely used in agriculture was proposed. The video key frame was obtained and the background modeling algorithm with strong adaptability to crop illumination was applied to realize real-time detection of the target, so as to construct the crop pixel model. Finally, the proposed algorithm and the constructed model are applied to the processing of shadows of agricultural plant video images for experimental verification. The results show that in video frames 47, 194 and 258, the probable learning pixel classification method can be used to determine the shaded part of each frame, which can greatly improve the detection accuracy of crop shadows. The research in this paper shows that the probability learning pixel classification method can better enhance the shadow robustness and accuracy of crop video images.

The subject of this study is the pixel-by-pixel controlled classification of multichannel satellite images distorted by additive white Gaussian noise. The paper aim is to study the effectiveness of various methods of image classification in a wide range of signal-to-noise ratios; an F-measure is used as a criterion for recognition efficiency. It is a harmonic mean of accuracy and completeness: accuracy shows how much of the objects identified by the classifier as positive are positive; completeness shows how much of the positive objects were allocated by the classifier. Tasks: generate random valuesof the brightness of the noise components, ensuring their compliance with the accepted probabilistic model; implement the procedures of element-wise controlled classification according to the methods of support vectors, logistic regression, neural network based on a multilayer perceptron for images distorted by noise; evaluate and analyze the results of objects bezel-wise classification of noisy images; investigate the effect of noise variance on classification performance. The following results are obtained. Algorithms of pixel-by-pixel controlled classification are implemented. A comparative analysis of classification efficiency in noisy images is performed. Conclusions are drawn. It is shown that all classifiers provide the best results for classes that mainly correspond to areal objects (Water, Grass) while heterogeneous objects (Urban and, especially, Bushes) are recognized in the worst way; classifiers based on the support vector machine and logistic regression show low recognition accuracy of extended objects, such as a narrow river (that belongs to the wide class of "water"). The presence of noise in the image leads to a significant increase in the number of recognition errors, which mainly appear as isolated points on the selected segments, that is, incorrectly classified pixels. In this case, the best value of the classification quality indicator is achieved using neural networks based on a multilayer perceptron.

Made available in DSpace on 2020-12-11T00:48:55Z (GMT). No. of bitstreams: 0 Previous issue date: 2019-10-01 Pro-Reitoria de Pesquisa da UFMG Fundação de Amparo à Pesquisa do Estado de Minas Gerais (FAPEMIG) Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) Cedro Textil, Reserva Vellozia, Parque Nacional da Serra do Cipo Plant phenology studies rely on long-term monitoring of life cycles of plants. High-resolution unmanned aerial vehicles (UAVs) and near-surface technologies have been used for plant monitoring, demanding the creation of methods capable of locating, and identifying plant species through time and space. However, this is a challenging task given the high volume of data, the constant data missing from temporal dataset, the heterogeneity of temporal profiles, the variety of plant visual patterns, and the unclear definition of individuals' boundaries in plant communities. In this letter, we propose a novel method, suitable for phenological monitoring, based on convolutional networks (ConvNets) to perform spatio-temporal vegetation pixel classification on high-resolution images. We conducted a systematic evaluation using high-resolution vegetation image datasets associated with the Brazilian Cerrado biome. Experimental results show that the proposed approach is effective, overcoming other spatio-temporal pixel-classification strategies. Univ Fed Minas Gerais, Dept Comp Sci, BR-31270901 Belo Horizonte, MG, Brazil Univ Estadual Campinas, Inst Comp, BR-13083970 Campinas, SP, Brazil Univ Estadual Paulista, Inst Biociencias, BR-13506900 Sao Paulo, Brazil Univ Estadual Paulista, IGCE, BR-13506900 Sao Paulo, Brazil Univ Stirling, Biol & Environm Sci, Fac Nat Sci, Stirling FK9 4LA, Scotland Univ Estadual Paulista, Inst Biociencias, BR-13506900 Sao Paulo, Brazil Univ Estadual Paulista, IGCE, BR-13506900 Sao Paulo, Brazil FAPEMIG: APQ-00449-17 FAPESP: 2013/50155-0 FAPESP: 2013/50169-1 FAPESP: 2009/54208-6 FAPESP: 2016/26170-8 FAPESP: 2018/06918-3 CNPq: 424700/2018-2 CAPES: 001 CAPES: 88881.145912/2017-01 Cedro Textil, Reserva Vellozia, Parque Nacional da Serra do Cipo: PELD-CRSC-17

Supervised classification and target recognition of Hyperspectral images (HSI) is a challenging task due to high dimensionality and spectral mixing. Straightforward cognitive computation and target...

In geometallurgy, a process model operating at the mineral liberation level needs quantitative textural information about the ore. The utilization of this information within process modeling and simulation will increase the quality of the predictions. In this study, descriptors derived from color images and machine learning algorithms are used to group drill core intervals into textural classes and estimate mineral maps by automatic pixel classification. Different descriptors and classifiers are compared, based on their accuracy and capacity to be automated. Integration of the classifier approach with mineral processing simulation is also demonstrated. The quantification of textural information for mineral processing simulation introduced new tools towards an integrated information flow from the drill cores to a geometallurgical model. The approach has been verified by comparing traditional geological texture classification against the one obtained from automatic methods. The tested drill cores are sampled from a porphyry copper deposit located in Northern Sweden.

A novel PolSAR image speckle reduction algorithm based on a new definition of similarity coefficient is proposed in this paper. Pixels in image are firstly classified into three types by threshold segmentation which is calculated with the similarity features. Then, weighted filtering is applied on the pixels selected according to their types, power features and similarity properties. Experimental results with measured data collected by NASA/JPL AIRSAR system show that the proposed method is more effective than Lee Filter not only in speckle suppression but also in polarimetric properties and structure feature preservation.

As the demand for immersive media increases, the importance of depth image is growing. As we see from the conventional researches, the depth images in the multiview camera system have holes around the objects, and it degrades the quality of the immersive video. To solve this problem, we propose an algorithm to conceal the hole in the depth image where the characteristics of the texture image are utilized. In the proposed algorithm, the holes are filled with the values predicted by the classification and confidence of pixels. The simulation results show that the proposed algorithm outperforms other conventional methods significantly.

Locating and identifying the components mounted on a printed circuit board (PCB) based on machine vision is an important and challenging problem for automated PCB inspection and automated PCB recycling. In this paper, we propose a PCB semantic segmentation method based on depth images that segments and recognizes components in the PCB through pixel classification. The image training set for the PCB was automatically synthesized with graphic rendering. Based on a series of concentric circles centered at the given depth pixel, we extracted the depth difference features from the depth images in the training set to train a random forest pixel classifier. By using the constructed random forest pixel classifier, we performed semantic segmentation for the PCB to segment and recognize components in the PCB through pixel classification. Experiments on both synthetic and real test sets were conducted to verify the effectiveness of the proposed method. The experimental results demonstrate that our method can segment and recognize most of the components from a real depth image of the PCB. Our method is immune to illumination changes and can be implemented in parallel on a GPU.

The detection of objects has become a critical step to update ground cover information, and the availability of very high-resolution satellite images made us discover new classification methods that give us more details such as pixel classification. This study aims to explore the potential and performance of machine learning algorithms in poor urban conditions in order to show the power of the deep neural networks to detect objects and, more precisely, to detect roads. We propose a U-net architecture for road extraction from Massachusetts dataset. The results have been compared with different automatic classification learning algorithms. The results of the classification using U-net showed a high accuracy of 97.7%, more precise than all the other models, which is why it is the best method to solve classification tasks for objects detection in large-scale datasets.

Urban landscapes are characterized as the fastest changing areas on the planet. However, regularly monitoring of larger areas it is not feasible using UAVs or costly air borne data. In these situations, satellite data with a high temporal resolution and large field of view are more appropriate but suffer from the lower spatial resolution (deca-meters). In the present study we show that by using freely available Sentinel-2 data from the Copernicus program, we can extract anthropogenic features such as roads, railways and building footprints that are partly or completely on a sub-pixel level in this kind of data. Additionally, we propose a new metric for the evaluation of our methods on the sub-pixel objects. This metric measures the performance of the detection of an object while penalizing the false positive classification. Given that our training samples contain one class, we define two thresholds that represent the lower bound of accuracy for the object to be classified and the background. We thus avoid a good score in occasions where we classify correctly our object, but a wide area of the background has been included in our prediction. We investigate the performance of different deep-learning architectures for sub-pixel classification of the different infrastructure elements based on Sentinel-2 multispectral data and the labels derived from the UAV data. Our study area is located in the Rhone valley in Switzerland where very high-resolution UAV data was available from the University of Applied Sciences. Highly accurate labels for the respective classes were digitized in ArcGIS Pro and used as ground-truth for the Sentinel data. We trained different deep learning models based on state-of-the-art architectures for semantic segmentation, such as DeepLab and U-Net. Our approach focuses on the exploitation of the multi spectral information to increase the performance of the RGB channels. For that purpose, we make use of the NIR and SWIR 10m and 20m bands of the Sentinel-2 data. We investigate early and late fusion approaches and the behavior and contribution of each multi spectral band to improve the performance in comparison to only using the RGB channels. In the early fusion approach, we stack nine (RGB, NIR, SWIR) Sentinel-2 bands together, pass them from two convolutions followed by batch normalization and relu layers and then feed the tiles to DeepLab. In the late fusion approach, we create a CNN with two branches with the first branch processing the RGB channels and the second branch the NIR/SWIR bands. We use modified DeepLab layers for the two branches and then concatenate the outputs into a total output of 512 feature maps. We then reduce the dimensionality of the result into the finaloutput equal to the number of classes. The dimension reduction step happens in two convolution layers. We experiment on different settings for all of the mentioned architectures. In the best-case scenario, we achieve 89% overall accuracy. Moreover, we measure 60% building accuracy, streets accuracy 60%, railway accuracy 73%, river accuracy 92% and background accuracy 94%.

A post-classification processing technique for noisy multichannel image that includes the method of local spatial filtering of segmented images in pseudo colors and multilayer classification algorithm is proposed. The post-classification processing effectiveness is verified for satellite image degraded by additive white Gaussian noise. It is demonstrated that the presence of noise in remote sensing data complicates the recognition task in pixel-by-pixel classification leading to a large number of arising errors. It is shown that post-classification processing can significantly increase overall probability of correct recognition of objects in noisy images.

We present a novel method to assess the variations in protein expression and spatial heterogeneity of tumor biopsies with application in computational pathology. This was done using different antigen stains for each tissue section and proceeding with a complex image registration followed by a final step of color segmentation to detect the exact location of the proteins of interest. For proper assessment, the registration needs to be highly accurate for the careful study of the antigen patterns. However, accurate registration of histopathological images comes with three main problems: the high amount of artifacts due to the complex biopsy preparation, the size of the images, and the complexity of the local morphology. Our method manages to achieve an accurate registration of the tissue cuts and segmentation of the positive antigen areas.

11 pages; International audience; The present study proposes a workflow to extract from orthomosaics the enormous amount of dry stones used by past societies to construct funeral complexes in the Mongolian steppes. Several different machine learning algorithms for binary pixel classification (i.e. stone vs non-stone) were evaluated. Input features were extracted from high-resolution orthomosaics and digital elevation models (both derived from aerial imaging). Comparative analysis used two colour spaces (RGB and HSV), texture features (contrast, homogeneity and entropy raster maps), and the topographic position index, combined with nine supervised learning algorithms (nearest centroid, naive Bayes, k-nearest neighbours, logistic regression, linear and quadratic discriminant analyses, support vector machine, random forest, and artificial neural network). When features are processed together, excellent output maps, very close to or outperforming current standards in archaeology, are observed for almost all classifiers. The size of the training set can be drastically reduced (to ca. 300 samples) by majority voting, while maintaining performance at the highest level (about 99.5% for all performance scores). Note, however, that if the training set is inadequate or not fully representative, the classification results are poor. That said, the methods applied and tested here are extremely rapid. Extensive mapping, which would have been difficult with traditional, manual, or semi-automatic delineation of stones using a vector graphics editor, now becomes possible. This workflow generally surpasses pedestrian surveys using differential GPS or a total station.

When training a classifier the choice of loss function heavily influences the characteristics of the resulting model. The most commonly used loss function for classification is cross entropy. In image segmentation problems where each pixel is assigned to a particular class, overlap-based losses have recently been shown to improve classifier performance especially for datasets with an imbalanced class distribution. This is particularly relevant to segmentation because class imbalance mitigation strategies used in regular classification are often not applicable. Overlap-based losses, however, have different drawbacks. We are aiming at combining the upsides of different losses with a simple scheduling scheme during training while minimizing their downsides. Gradually transitioning from an overlap-based dice loss to cross entropy allows to reliably select a distinct minimum in the optimization landscape as a valuable alternative to results obtained from traditional unscheduled loss functions. We demonstrate the efficacy of our approach on different combinations of loss functions, datasets, and models.

As coral reefs continue to degrade globally due to climate change, considerable effort and investment is being put into coral restoration. The production of coral offspring via asexual and sexual reproduction are some of the proposed tools for restoring coral populations and will need to be delivered at scale. Simple, inexpensive, and high-throughput methods are therefore needed for rapid analysis of thousands of coral offspring. Here we develop a machine learning pipeline to rapidly and accurately measure three key indicators of coral juvenile fitness: survival, size, and color. Using machine learning, we classify pixels through an open-source, user-friendly interface to quickly identify and measure coral juveniles on two substrates (field deployed terracotta tiles and experimental, laboratory PVC plastic slides). The method’s ease of use and ability to be trained quickly and accurately using small training sets make it suitable for application with images of species of sexually produced corals without existing datasets. Our results show higher accuracy of survival for slides (94.6% accuracy with five training images) compared to field tiles measured over multiple months (March: 77.5%, June: 91.3%, October: 97.9% accuracy with 100 training images). When using fewer training images, accuracy of area measurements was also higher on slides (7.7% average size difference) compared to tiles (24.2% average size difference for October images). The pipeline was 36× faster than manual measurements. The slide images required fewer training images compared to tiles and we provided cut-off guidelines for training for both substrates. These results highlight the importance and power of incorporating high-throughput methods, substrate choice, image quality, and number of training images for measurement accuracy. This study demonstrates the utility of machine learning tools for scalable ecological studies and conservation practices to facilitate rapid management decisions for reef protection.

In the course of the latest COVID-19 flu epidemic, several projects have been carried out to test LD-based strategies for the helping diagnosis of lung diseases. Deeper learning (DL) has proven its effectiveness in radiography. Although the present study relies on CT scans, DL strategies for interpreting pulmonary ultrasound (LUS) images are being used in this article. In specific, we present a new, completely annotated LUS data collection obtained from multiple Italian institutions with labels showing the level of disease intensity in a shot, photo, and digit optimization mask. By using these data, we implement numerous profound models that deal with the related tasks of automated LUS image analysis. We introduce a new deeper network derived from Space Converter Networks, that continuously estimates the extreme disease score for an input frame and weakly controlled the location of pathological machines. We implement also a new approach for efficient video-level averaging of frames based on uninorms. Finally, we benchmark deep state-of-the-art models for estimating COVID-19 biomarker pixel classification. Experiment was conducted on the planned dataset show satisfactory results for all the tasks considered which will pave the way for potential DL studies for the diagnosis of LUS-based COVID-19.

Changing atmospheric conditions often result in a data distribution shift in remote sensing images for different dates and locations making it difficult to discriminate between various classes of interest. To alleviate this data shift issue, we introduce a novel supervised classification framework, called Classify-Normalize-Classify (CNC). The proposed scheme uses a two classifier approach where the first classifier performs a rough segmentation of the class of interest (COI) in the input image. Then, the median signal of the estimated COI regions is subtracted from all image pixels values to normalize them. Finally, the second classifier is applied to the normalized image to produce the refined COI segmentation. The proposed methodology was tested to detect deforestation using bitemporal Landsat 8 OLI images over the Amazon rainforest. The CNC framework compared favorably to benchmark masks of the PRODES program and state-of-the-art classifiers run on surface reflectance images provided by USGS.

Staff-line removal is an important preprocessing stage for most optical music recognition systems. Common procedures to solve this task involve image processing techniques. In contrast to these traditional methods based on hand-engineered transformations, the problem can also be approached as a classification task in which each pixel is labeled as either staff or symbol, so that only those that belong to symbols are kept in the image. In order to perform this classification, we propose the use of convolutional neural networks, which have demonstrated an outstanding performance in image retrieval tasks. The initial features of each pixel consist of a square patch from the input image centered at that pixel. The proposed network is trained by using a dataset which contains pairs of scores with and without the staff lines. Our results in both binary and grayscale images show that the proposed technique is very accurate, outperforming both other classifiers and the state-of-the-art strategies considered. In addition, several advantages of the presented methodology with respect to traditional procedures proposed so far are discussed. This work was supported by the Spanish Ministerio de Educación, Cultura y Deporte through a FPU Fellowship (Ref. AP2012–0939), the Spanish Ministerio de Economía y Competitividad through Project TIMuL (No. TIN2013-48152-C2-1-R supported by EU FEDER funds) and the Instituto Universitario de Investigación Informática (IUII) from the University of Alicante.

This paper presents a novel rule-based skin detection method that works in the YCbCr color space. The method is based on correlation rules that evaluate the combinations of chrominance values to identify the skin pixels in the YCb and YCr subspaces. The correlation rules depend on the shape and size of dynamically generated skin color clusters, which are computed on a statistical basis in the YCb and YCr subspaces for each single image, and represent the areas that include most of the candidate skin pixels. Comparisons with six well-known rule-based methods in literature carried out on four publicly available databases show that the proposed method outperforms the others in terms of quantitative performance evaluation parameters. Moreover, the qualitative analysis shows that the method achieves satisfactory results also in critical scenarios, including severe variations in illumination conditions.

Methods for reticulocyte identification and counting based on digital image processing technology are rare. In this paper, we proposed a pixel-based reticulocyte identification method for blood micrographs. This approach not only addresses slowness of manual methods but also largely alleviates the susceptibility of flow-based methods to nucleic acid. The key is to extract pixel-level features. One color feature is extracted from three color spaces, namely RGB, HSI, and LUV. Three texture features called Gabor features, the gray-level co-occurrence matrix, and local contrast pattern are extracted from the Cr channel of the YCbCr color space, forming six features. Subsequently, the recognition effects of each combination of color and texture were compared, and the combination of Gabor texture features and LUV color features was selected. Then, a support vector machine (SVM) classifier was used to classify the pixel-level features, and the RNA-staining area was detected. Based on the location, number, and area of the region, whether the target cell is reticulocyte can be determined. The precision of this method for reticulocytes was 98.4%, recall was 98.0%, and the F1 measure was 0.982, indicating its usefulness for automation equipment.

Convolutional neural networks have gradually dominated the field of image semantic segmentation, and have achieved good results in 2D image semantic segmentation tasks. However, the 2D semantic segmentation algorithm based on CNN is still unsatisfactory in many complex scenarios, such as indoor scenes. Fortunately, advances in depth sensor technology have made it easy to obtain depth information, which carries rich geometric structure information. In order to effectively embed the depth map into the convolutional neural network, this paper introduces the dual encoder fusion network framework to fully exploit the geometric features. For the problem of weakening the local pixel classification ability of the dual encoder fusion network, this paper introduces global convolutional network (GCN), which is based on the large kernel idea, to improve the performance of dual encoder fusion network. Extensive experiments on the NYU v2 dataset demonstrate that the two-encoder fusion network based on global convolution network has much better precision than the original fusion network, and the classification ability for local pixels is stronger.

Researchers proposed various visual based methods for estimating the fruit quantity and performing qualitative analysis, they used ariel and ground vehicles to capture the fruit images in orchards. Fruit yield estimation is a challenging task with environmental noise such as illumination changes, color variation, overlapped fruits, cluttered environment, and branches or leaves shading. In this paper, we proposed a learning free fast visual based method to correctly count the apple fruits tightly overlapped in a complex outdoor orchard environment. We first carefully build the color based HS model to perform the color based segmentation. This step extracts the apple fruits from the complex orchard background and produces the blobs representing apples along with the additional noisy regions. We used the fine tuned morphological operators to refine the blobs received from the previous step and remove the noisy regions fol-lowed by the Gaussian smoothing. Finally we treated the circular shaped blobs with Hough Transform algorithm to calculate the center coordinates of each apple edge and the method correctly locates the apples in the images. The results ensures the proposed algorithm successfully detects and count apple fruits in the images captured from apple orchard and outperforms the standard state of the art contoured based method.

Semantic segmentation using fully convolutional networks has quickly become a popular solution as they provide very accurate per pixel classification. However, the implementation of deconvolutional layers and their mechanics differ greatly to those of patch based segmentation using convolutional neural networks. Both techniques have been used for road segmentation from satellite imagery but never compared. Thus we investigate the difference between fully connected and deconvolutional layers and provide an interpretation as to the correlation and differences between each methodology for road segmentation from satellite imagery.

This work presents a novel approach to tackle the music staff removal. This task is devoted to removing the staff lines from an image of a music score while maintaining the symbol information. It represents a key step in the performance of most optical music recognition systems. In the literature, staff removal is usually solved by means of image processing procedures based on the intrinsics of music scores. However, we propose to model the problem as a supervised learning classification task. Surprisingly, although there is a strong background and a vast amount of research concerning machine learning, the classification approach has remained unexplored for this purpose. In this context, each foreground pixel is labelled as either staff or symbol. We use pairs of scores with and without staff lines to train classification algorithms. We test our proposal with several well-known classification techniques. Moreover, in our experiments no attempt of tuning the classification algorithms has been made, but the parameters were set to the default setting provided by the classification software libraries. The aim of this choice is to show that, even with this straightforward procedure, results are competitive with state-of-the-art algorithms. In addition, we also discuss several advantages of this approach for which conventional methods are not applicable such as its high adaptability to any type of music score. This work has been supported by the Spanish Ministerio de Educación, Cultura y Deporte through a FPU Fellowship (Ref. AP2012–0939) and the Spanish Ministerio de Economía y Competitividad through Project TIMuL (No. TIN2013-48152-C2-1-R, supported by UE FEDER funds).

One of the fundamental challenges in the design of perception systems for autonomous vehicles is validating the performance of each algorithm under a comprehensive variety of operating conditions. In the case of vision-based semantic segmentation, there are known issues when encountering new scenarios that are sufficiently different to the training data. In addition, even small variations in environmental conditions such as illumination and precipitation can affect the classification performance of the segmentation model. Given the reliance on visual information, these effects often translate into poor semantic pixel classification which can potentially lead to catastrophic consequences when driving autonomously. This paper presents a novel method for analysing the robustness of semantic segmentation models and provides a number of metrics to evaluate the classification performance over a variety of environmental conditions. The process incorporates an additional sensor (lidar) to automate the process, eliminating the need for labour-intensive hand labelling of validation data. The system integrity can be monitored as the performance of the vision sensors are validated against a different sensor modality. This is necessary for detecting failures that are inherent to vision technology. Experimental results are presented based on multiple datasets collected at different times of the year with different environmental conditions. These results show that the semantic segmentation performance varies depending on the weather, camera parameters, existence of shadows, etc.. The results also demonstrate how the metrics can be used to compare and validate the performance after making improvements to a model, and compare the performance of different networks.

In this paper, we propose a novel method for pixel classification of remotely sensed images. The proposed method exploits the spatial information of image pixels using morphological profiles produced by structuring elements of different sizes and shapes. Morphological profiles produced by multiple structuring elements are combined into a single feature by decimal coding. The advantage of proposed feature is that it can effectively utilize the potential of multiple morphological profiles without increasing the complexity of feature space. The proposed approach was tested on remotely sensed images with known ground truths, and performance was improved up to 27 % in the overall accuracy results over existing techniques.

This research article addresses the problem of land-cover classification from the multi-spectral remotely sensed images using a novel self-training based semi-supervised learning (SSL) technique. The proposed system, instead of using a single classifier, builds an ensemble of classifiers with the hope that the ensemble system will have a lesser generalization error than any of its members. Each component classifier is trained independently using the proposed self-training approach on different training sub-sets and their predictions on test samples are combined using a new instant run-off voting (IRV) based classifier combination method. Each training subset consists of a few labeled samples and a comparatively large number of unlabeled data items. The cluster-and-label method for self-training has been adopted here considering the support vector machines (SVM) as the supervised learner and minimum spanning tree based clustering. A normalized histogram intersection kernel has been proposed which has shown to outperform the state-of-the-art kernel functions in terms of the SVM generalization accuracy. A novel cluster validity index specifically for graph based clustering has also been introduced to access the quality of the clustering. The proposed training method has the advantage that it can find classes for which the labeled training data are unavailable initially. A number of multi-spectral images have been considered for the experimental evaluation and a comparison with some of the well-established techniques from the literature has confirmed the superiority of the proposed classifier system.

Digital staining based on Mueller matrix measurements and their derivatives was investigated. Mueller matrix imaging was performed at the microscopic level on gastric tissue sections. Full Mueller matrices (4 × 4) were reconstructed using recorded images, followed by the extraction of polarization parameters. The most effective parameters and their combinations were extracted from Mueller matrix elements, principal component scores and polarization parameters respectively to classify samples into three categories - i.e. cancer, dysplasia and intestinal metaplasia/normal glands for various regions of interest sizes. It was observed that two-step classification yielded higher classification accuracy than the traditional one-step classification and that pixel classification based on Mueller matrix elements yielded higher accuracy than that based on polarization parameters and derived principal components. Moreover, Mueller matrix images with a lower spatial resolution generated higher classification accuracy but those with a higher spatial resolution revealed more morphological details.ns. The original stained image (top) and the digital staining image (bottom).

We present a reversible data hiding method for encrypted images that guarantees reversibility, i.e., the exact recovery of secret data and cover image by the receiver. This work is based on Zhang's reversible data hiding in encrypted images. Spatial correlation of pixels in the cover image is exploited for data hiding. In the proposed method, the sender proactively chooses suitable blocks in the cover image to hide secret data. The stego image is then encrypted so that privacy of cover image as well as secret data is protected. Zhang's method attains reversibility only with sufficiently large block size, thereby sacrificing payload capacity. The proposed method achieves reversibility at all block sizes, while significantly improving the embedding capacity over many of the state-of-the-art methods.

Hyperspectral data provide detailed information about the spectral properties of an observed scene. Although hyperspectral images contain much information, the reduction of dimensionality of these data is sometimes necessary to minimize their processing complexity. Band selection techniques are ways to perform dimensionality reduction. These techniques consist in choosing some spectral bands that best represent information contained in the original data. In this paper, a hyperspectral band selection approach is proposed. This approach is a sequential clustering-based method. Experiments, based on real hyperspectral data sets, are conducted to assess the performance of the proposed approach and other literature methods. The evaluation is performed by using a supervised pixel classification. The obtained accuracies show that the proposed approach gives satisfactory results and outperforms the tested literature methods.

Although singular spectrum analysis (SSA) has been successfully applied for data classification in hyperspectral remote sensing, it suffers from extremely high computational cost, especially for 2D-SSA. As a result, a fast implementation of 2D-SSA namely F-2D-SSA is presented in this paper, where the computational complexity has been significantly reduced with a rate up to 60%. From comprehensive experiments undertaken, the effectiveness of F-2D-SSA is validated producing a similar high-level of accuracy in pixel classification using support vector machine (SVM) classifier, yet with a much reduced complexity in comparison to conventional 2D-SSA. Therefore, the introduction and evaluation of F-2D-SSA completes a series of studies focused on SSA, where in this particular research, the reduction in computational complexity leads to potential applications in mobile and embedded devices such as airborne or satellite platforms.

The development of an automatic telemedicine system for screening and grading of Diabetic Retinopathy depends (DR) on reliable detection of abnormalities in retinal fundus images. The abnormalities in retinal fundus images due to DR are lesions which includes Micro aneurysms and Haemorrhages. An automatic detection of abnormalities like Micro aneurysm and Haemorrhages in colour fundus images is important to grade the diabetes level of patients in an automated system. The proposed method to detect the abnormalities in retinal fundus image is based on dynamic shape features and SVM classifier. Red lesion detected using two important aids. The first is candidate detection based on pixel classification. Using this technique vasculature and red lesions are separated from the back ground of the image. After removal of the connected vasculature the remaining objects are considered possible red lesions. Second the detected candidates are used to train the Support Vector Machine (SVM) classifier. The abnormalities detection accuracy of the proposed method is about 90% for diartedb1 database and 80% for stare database by cross training.

In recent years the interest of biomedical and computer vision communities in acquisition and analysis of epidermal images increased because melanoma is one of the deadliest form of skin cancer and its early identification could save lives reducing unnecessary medical treatments. User-friendly automatic tools can be very useful for physicians and dermatologists in fact high-resolution images and their annotated data, combined with analysis pipelines and machine learning techniques, represent the base to develop intelligent and proactive diagnostic systems. In this work we present two skin lesion detection pipelines on dermoscopic medical images, by exploiting standard techniques combined with workarounds that improve results; moreover to highlight the performance we consider a set of metrics combined with pixel labeling and classification. A preliminary but functional evaluation phase has been conducted with a sub-set of hard-to-treat images, in order to check which proposed detection pipeline reaches the best results.

The fuzzy c-means (FCM) clustering is an unsupervised clustering method, which has been widely used in image segmentation. In this paper, a spatially constrained fuzzy c-means clustering algorithm for image segmentation is proposed to overcome the sensitivity of the FCM clustering algorithm to noises and other imaging artifacts. Firstly, the local prior probabilities of pixel classification are defined according to the fuzzy membership function values of neighbouring pixels, and then those local prior probabilities are incorporated into the objective function of the standard FCM. Thus, the local spatial information embedded in the image is incorporated into the FCM algorithm. Experimental results on the synthetic and real images are given to demonstrate the robustness and validity of the proposed algorithm.

This study proposes a method to combine the k-Nearest Neighbor (k-NN) algorithm and the Support Vector Machine (SVM) method to increase the image annotation accuracy. Image annotation is widely employed in domains such as web image classification, search, military, and biomedicine. Although the traditional Border/Interior pixel Classification (BIC) features are very efficient and compact when applied to image annotation to capture color, shape, and texture information, the color space histogram utilization rates are not balanced. The experiment results show that the Hilbert-scan method and the One-pass Partitioning Method (OPM) can effectively overcome the imbalance problem.

Summary State-of-the-art light and electron microscopes are capable of acquiring large image datasets, but quantitatively evaluating the data often involves manually annotating structures of interest. This process is time-consuming and often a major bottleneck in the evaluation pipeline. To overcome this problem, we have introduced the Trainable Weka Segmentation (TWS), a machine learning tool that leverages a limited number of manual annotations in order to train a classifier and segment the remaining data automatically. In addition, TWS can provide unsupervised segmentation learning schemes (clustering) and can be customized to employ user-designed image features or classifiers. Availability and implementation TWS is distributed as open-source software as part of the Fiji image processing distribution of ImageJ at http://imagej.net/Trainable_Weka_Segmentation . Contact ignacio.arganda@ehu.eus. Supplementary information Supplementary data are available at Bioinformatics online.

There is a requirement for optic disc dataset with manual segmentation delimited by ophthalmologist and ground truth to compare segmented image obtained by pixel classification method with its ground truth. We tried to address this requirement by collecting some retinal images from local eye clinic and made optic disc marked by ophthalmologist. A widely used online dataset images are also marked by ophthalmologist for optic disc. The ground truths for those images are prepared. This paper presents optic disc localization and segmentation of retinal images obtained from local as well as online public dataset by Artificial Neural network, the results of which are quite satisfactory.

In this paper, we investigate the use of coarse image features for predicting class label distributions at a finer scale. The major contributions of this work are 1) use of coarse image features to improve the optimization formulation of conventional rank based approaches 2) use of inter class compatibility information from coarse images to refine the predicted target distribution 3) an enhanced unsupervised variogram based sub-pixel mapping approach 4) inclusion of abundance estimation uncertainty in the unmixing process. The proposed modifications on rank based and variogram based approaches have produced an accuracy improvement of 10–15%. The sensitivities of these approaches towards tunable parameters are also analyzed.

In this article, we present a method of nonmarked road following that is based on images coming from an onboard monochromatic camera. The principle is based first on a segmentation stage that makes it possible to locate the road area in the image, managing, if possible, the shadows on the roadway. The method is original since the algorithm must be running day as well as night (infrared camera) so it does not use color images. Furthermore, a single constant threshold is used whatever the analyzed sequence. Then, a localization stage estimates the vehicle’s location on the roadway. The estimate of the parameters L (road width) and α (camera inclination angle) (assumed known and constant for certain existing approaches) ensures a greater robustness of the other estimated parameters. Finally, a filtering stage is applied onto the previous data and predicts the position of the vehicle in the next image. Results are shown for each stage on both a normal nonmarked road and a forest lane sequence. The computational times are very low and will permit a real-time implementation on an experimental vehicle.

Cluster headache diagnosis is an open problem, whose signs are often noticeable in terms of changes in texture, color, or clarity of the eye of the induced pain side. The color difference is sometimes visible by the naked eye, hence it is close to the diagnosis of the cluster headache. Preliminary diagnosis tests on patients with cluster headache have shown that all of them have hypo-pigmented iris in the symptomatic side. We present here a color quantization method which can document the pigmentation differences of the iris, aiming to develop the first step towards an automatic diagnosis detection system. To do this, we used a statistical learning approach, specifically a Support Vector Classifier, to detect differences between iris pigmentation by using the error probability as a surrogate of color differences between the pigmentation in both eyes. The classification performance was compared when considering several features from different color spaces, in such a way that the error probability provided by the classifier when comparing the iris color of both eyes of the same patient provides a quantitative measure of the headache diagnosis. Systematic tests were performed on a database with images of 11 patients (seven patients with cluster headache and four control subjects). We can conclude from the present work that the study of the iris color features through statistical learning emerges as a technique of interest in the study of disorders affecting the sympathetic system.