GPU-Accelerated Histogram Generation on Smart-Phone and Webbrowser

Histogram is a critical component of algorithms in image processing to count tonal percentage. The performance of generating histogram sequentially does not satisfy the demand of realtime applications on smart phone and webbrowser. This paper proposes a two-pixel voting scheme (2PVS) for histogram generation on GPU. Compared with previous methods, the scale of problem can be cut down by a half using 2PVS. Every two adjacent pixels are considered as one object to be voted into a bin of histogram, followed by a recursive texture reduction process. We implement this method with graphics interface, which is compatible with embedded device and webbrowser. Experiments show that our method runs 0.3 to 1.9 times faster than the baseline method on smartphone while 1.2 to 2.6 times faster on webbrowser.