Cluster

Project summary:

In this paper, we propose a novel local descriptor, named quaternionic local ranking binary pattern (QLRBP), for color images. QLRBP is based on QR of the color image and treats all color channels in a holistic way. It not only reveals the stereo characteristics of the original color image but also possesses robustness to different variations. After analyzing LBP and its improvements, we come out a new perspective that these methods can be converted into a scheme of designing a proper ranking function to determine the ordering between two pixels. In QLRBP, to propose a quaternionic ranking function, a Clifford translation of quaternion (CTQ) is applied to QR of the color image such that the phase of the CTQ result is able to determine the order of two color pixels and its physical meaning is easily interpreted. The LBP coding is then applied to the phase image to obtain a QLRBP coding image, from which we extract the dense local histograms as the local image descriptor. Moreover, the properties of CTQ is also deeply studied. The proposed QLRBP is evaluated by the person reidentification and face recognition problems. Comparison results show the effectiveness of QLRBP.

Figure 1. Example of QLRBP coding. (a) The input 3*3 image patches for red, green, and blue channels respectively. (b) The QRs of all color pixels. (c) The CTQ results of all pixels. The reference quaternion \dot{p}= 0.8815i+0.4489j +0.1463k. (d) The weighted L1 phases of the CTQ results in (c). All weights are set to 1. (e) The ranking results between the center pixel and its surrounding ones. (f) The binary string obtained from the ranking results.

Experimental results:

Figure 2. CMC results of different feature extraction methods on the i-LIDS MCTS.

Reference:

Rushi Lan, Yicong Zhou*, and Yuan Yan Tang, “Quaternionic Local Ranking Binary Pattern: A Local Descriptor of Color Images,” IEEE Transactions on Image Processing, vol. 25, no. 2, pp. 566–579, 2016.