Visual Object Matching Based on Gradient ICA Feature

Pixel point gradient features represent much of the intrinsic structures of an image and can be used to the description of machine vision object. By ICA technique, pixel gradient data can be projected from a high-dimensional space to a lower-dimensional space, which reduce the redundancy with no image segment based on threshold. A method of visual object matching based on gradient ICA feature is provided in the paper. By training, the gradient ICA features description of both template and object can be acquired. And normalized cross correlation of the gradient ICA feature is adopted as the similar measure for the matching. Matching search can be easily realized from coarse to fine. Matching pulse correlation coefficient is high, and when there is non-uniform illumination or noise, the object can also be clearly recognized, which has be verified by the experiments.