A New Compressor for Measuring Distances among Images

Ideally, we would like to have a measure of similarity between images that did not require a feature selection and extraction step. In theory, this can be attained using Kolmogorov complexity concepts. In practice, because the Kolmogorov complexity of a digital object cannot be computed, one has to rely on appropriate approximations, the most successful being based on data compression. The application of these ideas to images has been more difficult than to some other areas. In this paper, we suggest a new distance and compare it with two others, showing some of their relative advantages and disadvantages, hoping to contribute to the advance of this promising line of research.