Region edge histogram: A new feature for region-based tracking

We present a new histogram feature for region-based tracking, which we refer as region edge histogram (REH). Firstly, we introduce some new concepts such as bin region and edge complexity. A bin region consists of the pixels falling in the corresponding histogram bin, and edge complexity denotes a ratio of the circumference to the area of bin region. Then, we propose our feature based on conventional histogram and the edge complexity of its bin region. We confirm that a higher edge complexity shows a more complicated space distribution to bin region. Finally, we derive a compact form of REH and discuss its accelerated computing method based on integral histogram. We design a region-based tracker using exhaustive search method to compare the performances of conventional histogram, weighted histogram, spatiogram and REH. Experiments show that REH obtains the lowest tracking errors and owns a relative high efficiency to conventional histogram, which is much higher than that of the other two.