Robust Object Tracking using Log-Gabor Filters and Color Histogram

Texture analysis is a major challenge for computer vision, since image textures process are widely used to perform tasks such as segmentation and object tracking. The performance of the tracking algorithm relies heavily on the accuracy of relevant information extracted from the object to track. Our hypothesis is that by adding log-Gabor filter to color features, and then embedded it to the mean shift framework, tracking performance will enhance outstandingly. Apart from the classical color histogram, and in order to expedite the extraction of parameters of the target model in complex situations, the Log-Gabor filter is employed jointly with color feature in this paper. The design method and implementation scheme of Log-Gabor filter and its feature encoding for texture extraction are described in detail. Compared with the LBP texture-color histogram based algorithms, our proposed method extracts effectively pertinent information in the target region, which describe better and characterize more robustly the target. Experimental results show that the proposed Log-Gabor filtering with color feature method, improves substantially the tracking efficiency and accuracy with fewer mean shift iterations than the tracking algorithm using LBP texture.