Font Size: a A A

A Study On Face Recognition Based On Compressive Sensing

Posted on:2016-10-01Degree:DoctorType:Dissertation
Country:ChinaCandidate:G T ChengFull Text:PDF
GTID:1108330485955111Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
Compressive sensing theory is one of the international hot topics in signal process-ing field in recent years, and has penetrated into mathematics and many engineering fields. Sparse representation based face recognition is one of the representative applications of compressive sensing theory. It assumes that a testing image can be sparsely represented by the whole training images, and the sparse representation coefficient can be solved by the convex optimization technique. The final classification is accomplished by evaluating the minimum reconstructed residuals. It provides a new research perspective for the study of face recognition problem, and has attracted wide attention around the world. Derived from face recognition based on sparse representation, how to improve the robustness and efficiency of existing algorithms was discussed, and the effective face recognition algo-rithms were proposed. The main research works were shown as follows:1. A face recognition algorithm based on spare representation with 2D feature ma-trix was proposed. The original 2D face images were transformed into 2D Fisherface feature matrix, the spare representation of testing face was directly solved with this 2D matrix, and the final classification was accomplished by evaluating the minimum recon-structed residuals. Different from the previous face recognition algorithms, the whole process did not transformed 2D matrix into 1D vector. The proposed method not only retained relevance information between elements in the original 2D matrix, but also im-proved the complexity of solving the l1-norm minimization problem. For face image with contiguous occlusion, each image was partitioned into some blocks, and a new rule com-bining sparsity and reconstruction residual was defined to discard the occluded blocks, the final result was aggregated by voting the classification results of the valid individual blocks. The experimental results illustrated that the proposed algorithm outperformed the existing methods in term of accuracy and efficiency.2. In order to address the challenges that both the training and testing images were contaminated by random pixels corruption, occlusion, and disguise, a robust face recog-nition algorithm based on two-stage sparse representation was proposed. Specifically, noises in the training images were eliminated by low-rank matrix recovery. Then, by exploiting the first-stage sparse representation computed by solving a new extended l1-minimization problem, noises in the testing image were successfully removed. After the elimination noises in all face images, feature extraction techniques that are more dis-criminative but sensitive to noises were effectively performed on the reconstructed clean images, and the final classification was accomplished by utilizing the second-stage s-parse representation obtained by solving the reduced l1-minimization problem in a low-dimensional feature space. Feature extraction techniques exhibited good performance due to their being performed on the denoising face images. On the one hand, the proposed algorithm was robust to feature dimensions. Its performance did not vary obviously with the decrease of feature dimensions. On the other hand, the proposed algorithm was robust to noises disturbance level. With the increase of noise level, the recognition rate of the proposed algorithm degraded slowly.3. Aiming at improving the robustness of face recognition algorithms based on im-age set, the problem that the query image set suffered serious disturbances was discussed, and a robust and efficient image set face recognition method based on low rank and collab-orative representation was proposed. By introducing a Frobenius norm term, an extended low rank representation model was firstly developed to remove all possible disturbances from the query set and reconstruct the rank-one query set. To improve the efficiency, a compact and incoherent dictionary for the large gallery set was learned, and the closed form solutions for both the dictionary and the coding coefficient were straightway derived. The final classification was performed by using any frame in the reconstructed query set instead of using the whole set, which further improved the running efficiency. The exper-imental results exhibited that the proposed algorithm can achieve the more satisfactory robustness and efficiency that its competitors.
Keywords/Search Tags:Face recognition, Compressive sensing, Sparse representation, Low rank recovery, Robustness, Efficiency
PDF Full Text Request
Related items