Adaptive Patch Quantization for Histogram-Based Visual Tracking

Color histogram-based trackers have shown favorable performance recently. However, most color histogram-based trackers employ pixel-wise information, incapable of distinguishing objects from backgrounds robustly. In this paper, we propose an adaptive patch quantization approach for histogram-based visual tracking. We first exploit neighboring pixels in the form of local patches to improve the discrimination between objects and backgrounds. Then we propose an adaptive quantization strategy for quantization space update and histogram adjustment to avoid model drifting. We further exploit a novel localization technique based on adaptive segmentation to improve the localization accuracy. The experimental results demonstrate that the proposed method performs superiorly against several state-of-the-art algorithms.