Selective Color Mixture Estimation for Image Matching

摘要

Similarity estimation between images or an image and regions on images is still an open and interesting research theme. Several applications of image similarity estimation exists and more can be devised: searching and indexing of image databases, subimage template matching, applications in robot and computer vision, location of visual information on the World Wide Web, etc. If the images are in color, this additional information must be considered carefully since slightly variations on the color components may give different similarity indexes between images. A color similarity index between several images or one image and several regions of it is also sensitive to noise, small outliers and regions where colors are mixed. This paper presents a method for extraction of color features of images that works as a color mixture estimation, based on the selective attention filter, which is able to eliminate color noise from the images. The extracted features represent the most frequent colors in the image. Morphological features are not considered. The filtering and feature extraction will be executed on a small set of test images. Results of the experiments and discussion will be presented.