Fig. 6 summarizes experimental results on image matching and panorama stitching by setting the proportions of “unimportant” subregions in our algorithm. In this case we are setting the proportion of subregion to be categorized as regions with dense textures. Textureless regions are classified automatically. We also set other parameters so that the system maintains the false positive rate of lower than 3 percent. In Fig. 6(a), we tested image matching using three groups of 200 test images each, with slightly different image backgrounds in each group.