sparse coding

Due to the growing increase of generated images via cameras and various instruments, image processing has found an important role in most of practical usages including medical, security and driving. However, most of the available models has no considerable performance and in some usages the amount of error is very effective. The main cause of this failure in most of available models is the distribution mismatch across the source and target domains. In fact, the made model has no generalization to test data with different properties and distribution compared to the source data, and its performance degrades dramatically to face with new data. In this paper, we propose a novel approach entitled Sparse coding and ADAptive classification (SADA) which is robust against data drift across domains. The proposed model reduces the distribution difference across domains via generating a common subspace between the source and target domains and increases the performance of model. Also, SADA reduces the distribution mismatch across domains via the selection of the source samples which are related to target samples. Moreover, SADA adapts the model parameters to build an adaptive model to encounter with data drift. Our variety of experiments demonstrate that the proposed approach outperforms all stat-of-the-art domain adaptation methods.

Due to the rapid growth of information technology and exponential increasing of information the need for more and more storage capacity and efficiency has increased. Image compression is an important tool to reduce the redundancy of images data in order to be able to store or transmit them in an efficient manner. When images are limited to a specific and limited family of images like MRI databases of a hospital or facial image database of a university or an organization or fingerprint image databases, this limitation increases the total spatial redundancy. Thus, efficient storage of such images is beneficial, and their compression becomes an appealing application, and this urges algorithms specially tailored for the task of content base image compression to surpass general purpose compression algorithms. The facial images, due to their wide application as the most common images in the organizations and companies are more considerable for image compression. In this paper a new image compression scheme using sparse coding and RLS-DLA redundant dictionary learning is proposed that can be used for compressing of face image databases. In the proposed method, several dictionaries are exploited adaptively based on the required image quality to enhance the overall rate-distortion. The simulation results show that this scheme outperforms the state-of-art algorithms like JPEG2000 by about 0.5 to 1.2 dB for reconstructed images PSNR.

Domain adaptation can transfer knowledge from a training set (source domain) to a test set (target domain), promoting the performance of the model learned from the training set. In addition, sparse coding makes the learned model more succinct and easy to manipulate. However, the existence of the distribution mismatch across the source and target domains reduce the performance of model. In this paper, we propose an unsupervised domain adaptation model to minimize the prediction error of image classification. Sample reweighting is utilized to handle redundant and useless information of source data in the new representation. Moreover, the difference of the conditional distributions across the source and target domains is reduced along with the subspace alignment. Our proposed approach learns a sparse domain-invariant classifier in a latent subspace with preserving the structure of the input data. Extensive experiments demonstrate that our proposed approach shows 4.49% improvement in classification accuracy on real-world datasets compared to state-of-the-art machine learning and domain adaptation methods.

There are challenges such as depth perception and self-occlusion, in the field of 3D human pose estimation and reconstruction which obstructs precise estimation of body joints. In this paper, we first extract human pose by focusing on 2D ground-truth using sparse coding and. In the second approach, we use a learning-based Convolutional Neural Networks using sparse coding and a model based rectifier to extract the estimated pose. Pose estimation by proposedmethod has reduced the mean error of the reconstruction in comparison with the state of the artworks.

Texture and color are two important attributes for object recognition. Recently, quaternionic representation of color images have been used as an effective method for color image processing. Using such a representation, it is possible to consider the mutual interaction between different color channels. In the last decade, several quaternion operations like rotation, reflection, and Clifford translation have been developed. Such operators are able to extract shallow information from the color images. In this paper, we first propose a set of new quaternion operators called hybrid quaternionic operators, which can be produced by a cascade of several simple quaternionic operators. Such operators can extract deeper information from the color images. We then use such operators, and present a novel color texture classification method using the concept of sparse coding. Experimental results indicate that the proposed method outperforms several existing and popular methods.

The image observer usually wants to receive the message and the main subject of the image in the shortest time. Hence, assuming there is useful information in the salient regions, the human vision system unconsciously guides visual attention towards them. This assumption is not always correct in practice, and in some cases, salient regions merely cause visual distractions. Therefore, in different applications, a mechanism is needed to identify these regions. To prevent from distracting observer’s attention from the main subject, these regions are eliminated. Furthermore, neglecting these regions could be of considerable assistance to the methods that function base on salient regions recognition. So, in this paper, Based on the methods of the class imbalance challenge each segment of training images in the dataset is a partition to 9 classes according to the relevant mask in the dataset, that the number of each class is proportional to its disturbance intensity. Then, segment-based features are extracted and determining the class of each segment is determined according to WTPTSSR method, which is based on the Sparse Coding and Representation system.Finally, in order to precisely analyzing the proposed method and comparing it to other approaches, four analysis criteria with different performances are presented. According to results, despite being time-consuming, the proposed method has a higher accuracy than the previous ones.

Since the anomalies are unknown targets with low probabilities of occurrence which are significantly different from their neighbors, anomaly detection could be considered as one of the most important information extraction approaches from hyperspectral data. Various types of parametric and non-parametric algorithms have been developed in this area from the 1990's decade. Recently, sparse representation methods have been introduced and successfully accepted as a useful tool for anomaly detection based on the recovery of the majority of high-dimensional signals via a low-dimensional subspace through a dictionary of normalized signals called atoms. In other words, having a dictionary composed of bases denoting the background subspace enables the accurate recovery of background signals. Moreover, the presence of anomaly signals, assuming their deviation from the background subspace, will not have a precise estimation by the background dictionary. Hence the main idea of these anomaly detection methods is focused on evaluating recovery errors of signals by a dictionary that describes the background subspace. In such procedure, removing the atoms that describe the anomaly in the background dictionary can be considered as one of the essential actions. To this aim making diversity in the definition of spatial neighborhoods of spectral signals, as well as voting-based judgment in different situations of the spatial distribution could be proposed. In other words, by designing an optimized local dictionary, based on a local sliding window, the votes of each signal in terms of anomaly presence in each spatial neighborhood could be calculated with the aim of achieving better judgment. In this paper, a new anomaly detector for hyperspectral images is proposed based on simultaneous sparse representation using a new structured sliding window. The main contribution of this research is to improve the judgments about the anomaly presence probability using information collected during transition of the mentioned sliding window for each pixel under test. In this algorithm, each pixel experiences various spatial positions with respect to the neighbors through the transition of the sliding window. In each position, an optimized local background dictionary is molded using a well-known K-SVD method as an iterative process and the recovery error of sparse coding for each pixel under test is calculated using a simultaneous orthogonal matching pursuit algorithm (SOMP). So, the votes of each pixel in terms of the anomaly presence in each neighborhood are calculated and finally the variance of these estimated errors is considered as the anomaly detection criterion. The experimental results of the proposed method using four datasets (synthetic and real datasets) proved its higher performance compared to the GRX, LRX, CRD and BJSR detectors with an average efficiency improvement of about 9%. In addition automatic tuning of the proposed algorithm parameters (level of sparsity and the size of sliding window) and developing parallel processing techniques to improve the running time of this algorithm are the focus of our future research. It is notable that the implementation of this idea and its success showed that development of voting algorithms and the combination of the results could be considered as an efficient approach could also be utilized in other hyperspectral image processing algorithms.

One of the important topics of passive defense is threats detection and waning alarm. One of the most widely used methods in detection field is video data investigation in order to identify unknown targets and warning alarm. In order to evaluate a fast and high-precision technique, video summarization is presented. Also, during the past years, creation of digital videos has caused exponential growth of video content. To increase the high volume of video usability, a lot of researches have been done and video summarization has been proposed to quick view of large video collection and quick understanding of the content of video data. In the video summarization, pictures are selected as a representative of each scene to obtain a visual overview of whole video. Recently, new methods using sparse formulation are suggested for video summarization being more effective in video data summarization than other methods. In this paper, video summarization is presented as a sparse dictionary selection problem. For this purpose, using a new method based on sparse coding, have been able to improve video data summarization compared to other video summarization methods based on sparse or other coding. Finally, the results for the ground truth data collection and State of the art methods, shows improvement our claim in the video summary on proposed method.