Block-based fast integral histogram

Abstract

Integral Histogram is widely used among various realtime algorithms due to its computationally efficient facility to compute histograms from an image. However, its advantage is greatly diminished for algorithms requiring only few histograms since required initialization operation takes up majority of total running time. In this paper we propose a block based initialization that drastically reduces initialization time, a major weakness of integral histograms, and that uses interpolation methods to generate more reliable histogram from the data.