This thesis targets at developing local feature extraction technologies of color images in quaternion domain. Unlike traditional local features that are extracted from each color channel individually or from a specific color space, the proposed local features are based on the quaternionic representation (QR) of color images. QR encodes a color pixel using a quaternion number, thus being capable of handling all color channels directly in the quaternion domain and considering their relations. In this thesis, we propose three novel quaternionic local features using QR of color images from different perspectives. Comprehensive experimental results are provided to validate these proposed features with different applications. In summary, the main contributions of this thesis are exhibited as follows: • We are the first attempt to extend the well-known local binary pattern (LBP) into quaternion domain and propose a novel local feature called quaternionic local ranking binary pattern (QLRBP). First, a new understanding of LBP-based methods is provided in terms of ranking two pixels. Then, we derive a ranking function between two quaternions and apply the function to develop the QLRBP. QLRBP is more appropriate for color image classification because it considers both the characteristics of the neighboring pixels in a local patch and the interactions among different color components. • We propose a framework called quaternionic Weber local descriptors (QWLD) via transplating the Weber’s law in quaternion domain for local feature extraction. Based on this framework, two novel descriptors are developed as examples, namely quaternionic increment based Weber descriptor (QIWD) and quaternionic distance based Weber descriptor (QDWD). Extensive experiments are carried out to evaluate the proposed descriptors in multiple applications including face recognition, kinship verification, texture classification, and person reidentification. The comparison results show their effectiveness. • Inspired by the Michelson’s law, a framework named quaternion-Michelson descriptors (QMD) is proposed. Compared with the Weber’s law, the Michelson’s ii law stably measures the image contents from the viewpoint of human perception. QMD integrates both superiorities of Michelson contrast and QR. Using QMD framework, we further propose two novel quaternionic Michelson contrast binary pattern (QMCBP) descriptors from different perspectives. Experimental results demonstrate that the proposed framework and descriptors outperform several state-of-the-art local descriptors.