Integral image word length reduction using overlapping rectangular regions

Abstract

An integral image is a data table whose elements are the sums of pixel values taken from a source image. Using integral images, the total pixel value for any rectangular region in the source image can be calculated at high speeds. However, a large word length is required. In this study, we propose a word length reduction method for the integral image, which is based on the qualities of the integer addition operations. Using overlapping rectangular regions, we are able to reduce the word length and accurately calculate the total pixel values of the rectangular regions. This method can greatly reduce the word length when seeking the total pixel values of a large number of rectangular regions within the image.