Content-based image search for retrieval of images based on the similarity in their visual contents, such as color, texture, and shape, to a query image is an active research area due to its broad applications. Color, for example, provides powerful information for image search and classification. This dissertation investigates advanced image search techniques and presents new color descriptors for image search and classification and robust image enhancement and segmentation methods for iris recognition.

First, several new color descriptors have been developed for color image search. Specifically, a new oRGB-SIFT descriptor, which integrates the oRGB color space and the Scale-Invariant Feature Transform (SIFT), is proposed for image search and classification. The oRGB-SIFT descriptor is further integrated with other color SIFT features to produce the novel Color SIFT Fusion (CSF), the Color Grayscale SIFT Fusion (CGSF), and the CGSF+PHOG descriptors for image category search with applications to biometrics. Im­age classification is implemented using a novel EFM-KNN classifier, which combines the Enhanced Fisher Model (EFM) and the K Nearest Neighbor (KNN) decision rule. Exper­imental results on four large scale, grand challenge datasets have shown that the proposed oRGB-SIFT descriptor improves recognition performance upon other color SIFT descrip­tors, and the CSF, the CGSF, and the CGSF+PHOG descriptors perform better than the other color SIFT descriptors. The fusion of both Color SIFT descriptors (CSF) and Color Grayscale SIFT descriptor (CGSF) shows significant improvement in the classification per­formance, which indicates that various color-SIFT descriptors and grayscale-SIFT descrip­tor are not redundant for image search.

Second, four novel color Local Binary Pattern (LBP) descriptors are presented for scene image and image texture classification. Specifically, the oRGB-LBP descriptor is

derived in the oRGB color space. The other three color LBP descriptors, namely, the Color LBP Fusion (CLF), the Color Grayscale LBP Fusion (CGLF), and the CGLF+PHOG de­scriptors, are obtained by integrating the oRGB-LBP descriptor with some additional image features. Experimental results on three large scale, grand challenge datasets have shown that the proposed descriptors can improve scene image and image texture classification performance.

Finally, a new iris recognition method based on a robust iris segmentation approach is presented for improving iris recognition performance. The proposed robust iris seg­mentation approach applies power-law transformations for more accurate detection of the pupil region, which significantly reduces the candidate limbic boundary search space for increasing detection accuracy and efficiency. As the limbic circle, which has a center within a close range of the pupil center, is selectively detected, the eyelid detection approach leads to improved iris recognition performance. Experiments using the Iris Challenge Evaluation (ICE) database show the effectiveness of the proposed method.