Quaternion-Michelson Descriptor for Color Image Classification

Project Summary:

In this paper, we propose a novel framework called quaternion-Michelson descriptor (QMD) for color image classification. QMD derives the local descriptors of color images in the quaternion domain using the Michelson contrast. Compared with the Weber’s law, the Michelson contrast is a more stable measurement of the human perception of the image content. The QR-based Michelson contrast takes relations among different color channels into account to describe the perception of color images. Integrating the QR and Michelson contrast, QMD combines both of their advantages to provide a discriminative representation of the color image. Using the proposed QMD framework, as examples, we develop two quaternionic Michelson contrast binary pattern (QMCBP) descriptors. They extract the Michelson contrast of color images in the quaternion domain from different perspectives. A number of experiments are conducted to evaluate the proposed descriptors using different applications, and comparison results validate their effectiveness.

Lanrushi3

Fig. 1. A schematic picture of Quaternion-Michelson descriptor (QMD).

Experimental results:

Lanrushi4

Reference:

Rushi Lan, and Yicong Zhou*, "Quaternion-Michelson Descriptor for Color Image Classification," IEEE Transactions on Image Processing, vol. 25, no. 11, pp. 5281--5292, 2016.