Date of Award
Doctor of Philosophy in Computing Sciences - (Ph.D.)
Visual recognition is one of the most difficult and prevailing problems in computer vision and pattern recognition due to the challenges in understanding the semantics and contents of digital images. Two major components of a visual recognition system are discriminatory feature representation and efficient and accurate pattern classification. This dissertation therefore focuses on developing new learning methods for visual recognition.
Based on the conventional sparse representation, which shows its robustness for visual recognition problems, a series of new methods is proposed. Specifically, first, a new locally linear K nearest neighbor method, or LLK method, is presented. The LLK method derives a new representation, which is an approximation to the ideal representation, by optimizing an objective function based on a host of criteria for sparsity, locality, and reconstruction. The novel representation is further processed by two new classifiers, namely, an LLK based classifier (LLKc) and a locally linear nearest mean based classifier (LLNc), for visual recognition. The proposed classifiers are shown to connect to the Bayes decision rule for minimum error. Second, a new generative and discriminative sparse representation (GDSR) method is proposed by taking advantage of both a coarse modeling of the generative information and a modeling of the discriminative information. The proposed GDSR method integrates two new criteria, namely, a discriminative criterion and a generative criterion, into the conventional sparse representation criterion. A new generative and discriminative sparse representation based classification (GDSRc) method is then presented based on the derived new representation. Finally, a new Score space based multiple Metric Learning (SML) method is presented for a challenging visual recognition application, namely, recognizing kinship relations or kinship verification. The proposed SML method, which goes beyond the conventional Mahalanobis distance metric learning, not only learns the distance metric but also models the generative process of features by taking advantage of the score space. The SML method is optimized by solving a constrained, non-negative, and weighted variant of the sparse representation problem.
To assess the feasibility of the proposed new learning methods, several visual recognition tasks, such as face recognition, scene recognition, object recognition, computational fine art analysis, action recognition, fine grained recognition, as well as kinship verification are applied. The experimental results show that the proposed new learning methods achieve better performance than the other popular methods.
Liu, Qingfeng, "Investigation of new learning methods for visual recognition" (2017). Dissertations. 20.