Sparse coding in early visual representation: From specific properties to general principles

早期视觉表示的稀疏编码：从特殊性到一般原则

Neurocomputing, Volume 171, 1 January 2016, Pages 1085-1098

Abstract：In this paper, we examine the problem of learning sparse representations of visual patterns in the context of artificial and biological vision systems. There are a myriad of strategies for sparse coding that often result in similar feature properties for the learned feature set. Typically this results in a bank of Gabor-like or edge filters that are sensitive to a range of distinct angular and radial frequencies. The theory and experimentation that is presented in this paper serves to provide a better understanding of a number of specific properties related to low-level feature learning. This includes close examination of the role of phase pairing in complex cells, the role of depth information and its relationship to variation of intensity and chroma, and deriving hybrid features that borrow from both analytic forms and statistical methods. Together, these specific examples provide context for more general discussion of effective strategies for feature learning. In particular, we make the case that imposing additional constraints on mechanisms for feature learning inspired by biological vision systems can be useful in guiding constrained optimization towards convergence, or specific desirable computational properties for representation of visual input in artificial vision systems.

Irisrecognition based on sparse representation and k-nearest subspace with genetic algorithm

基于稀疏表示与K最近子空间以及遗传算法的Iris识别

Pattern Recognition Letters, Volume 73, 1 April 2016, Pages 13-18

Abstract：Iris recognition has become an important tool for human authentication. An efficient and robust iris recognition model based on sparse representation using compressive sensing and k-nearest subspace (segments) has been proposed; k-nearest subspace approach is used for short listing the classes to reduce the time. The shortlisted candidates are divided into sectors and the sparse recognition is applied to each sector. Three classifiers: k-nearest distance classifier, Sector based classifier and Cumulative Sparse Concentration Index (CSCI) based classifiers have been used. An additive function based classifier combination scheme has been adopted in which each classifier is associated with a weight. Genetic algorithm is used to learn the weight of each of the classifier. Results obtained on different databases show that the scheme is highly robust with FAR almost zero.

Sparse representation over learned dictionary forsymbol recognition

符号识别学习字典的稀疏表示

Signal Processing, Volume 125, August 2016, Pages 36-47

Abstract：In this paper we propose an original sparse vector model for symbol retrieval task. More specifically, we apply the K-SVD algorithm for learning a visual dictionary based on symbol descriptors locally computed around interest points. Results on benchmark datasets show that the obtained sparse representation is competitive related to state-of-the-art methods. Moreover, our sparse representation is invariant to rotation and scale transforms and also robust to degraded images and distorted symbols. Thereby, the learned visual dictionary is able to represent instances of unseen classes of symbols.

Shadow removal using sparse representation over local dictionaries

基于稀疏表示局部字典的阴影消除

Engineering Science and Technology, an International Journal, In Press, Corrected Proof, Available online 19 February 2016

Abstract：The presence of shadow in an image is a major problem associated with various visual processing applications such as object recognition, traffic surveillance and segmentation. In this paper, we introduce a method to remove the shadow from a real image using the morphological diversities of shadows and sparse representation. The proposed approach first generates an invariant image and further processing is applied to the invariant image. Here, shadow removal is formulated as a decomposition problem that uses separate local dictionaries for shadow and nonshadow parts, without using single global or fixed generic dictionary. These local dictionaries are constructed from the patches extracted from the residual of the image obtained after invariant image formation. Finally, non-iterative Morphological Component Analysis-based image decomposition using local dictionaries is performed to add the geometric component to the non-shadow part of the image so as to obtain shadow free version of the input image. The proposed approach of shadow removal works well for indoor and outdoor images, and the performance has been compared with previous methods and found to be better in terms of RMSE.

Noise robustness analysis of sparse representation based classification method for non-stationaryEEGsignal classification

基于非平稳EEG信号分类方法的稀疏表示噪音鲁棒性分析

Biomedical Signal Processing and Control, Volume 21, August 2015, Pages 8-18

Abstract：In the electroencephalogram (EEG)-based brain–computer interface (BCI) systems, classification is an important signal processing step to control external devices using brain activity. However, scalp-recorded EEG signals have inherent non-stationary characteristics; thus, the classification performance is deteriorated by changing the background activity of the EEG during the BCI experiment. Recently, the sparse representation based classification (SRC) method has shown a robust classification performance in many pattern recognition fields including BCI. In this study, we aim to analyze noise robustness of the SRC method to evaluate the capability of the SRC for non-stationary EEG signal classification. For this purpose, we generate noisy test signals by adding a noise source such as random Gaussian and scalp-recorded background noise into the original motor imagery based EEG signals. Using the noisy test signals and real online-experimental dataset, we compare the classification performance of the SRC and support vector machine (SVM). Furthermore, we analyze the unique classification mechanism of the SRC. We observed that the SRC method provided better classification accuracy and noise robustness compared with the SVM method. In addition, the SRC has an inherent adaptive classification mechanism that makes it suitable for time-varying EEG signal classification for online BCI systems.

Sparse molecular image representation

稀疏分子成像表示

Journal of Visual Communication and Image Representation, Volume 36, April 2016, Pages 213-228

Abstract：Sparsity-based models have proven to be very effective in most image processing applications. The notion of sparsity has recently been extended to structured sparsity models where not only the number of components but also their support is important. This paper goes one step further and proposes a new model where signals are composed of a small number of molecules, which are each linear combinations of a few elementary functions in a dictionary. Our model takes into account the energy on the signal components in addition to their support. We study our prior in detail and propose a novel algorithm for sparse coding that permits the appearance of signal dependent versions of the molecules. Our experiments prove the benefits of the new image model in various restoration tasks and confirm the effectiveness of priors that extend sparsity in flexible ways especially in case of inverse problems with low quality data.