An Effective Color Addition to Feature Detection and Description for Book Spine Image Matching
Figure 5
The source image in each channel is independently convolved with a Gaussian kernel at differing values of , and then adjacent images are differenced to produce a Difference of Gaussians. The channel DoG is shown in (b), the in (c), and the in (d). Different strong edges are visible in each channel. A grayscale feature detector would only find those visible in the channel and miss the very obvious features in the chrominance channels.