Let us summarize the image difference algorithm now. The integration of the contrast sensitivity function can be precomputed and stored in an array. There are two images (let us assume they are of equal size), both contain r, g, and b values for each pixel. First the r, g, and b values are transformed to CIE XYZ values, and summed area tables are built for X, Y, and Z values for both images (6 tables are built). The Halton series can be precomputed or easily computed on the fly. The size of a rectangle is determined, and its orientation and position in the images using Halton series. Average X, Y, and Z of the rectangles are computed using (8.5) and colors are converted to the CIE LUV color space. Now the color difference is computed using (8.11) and this difference is added to the total distance. At the end, the total distance is divided by the number of rectangles and gives us the difference between two images. Algorithm 3 shows the pseudo code of the whole process.
Algorithm 3