I have coauthored, with Jean-Paul Delahaye and Cedric Gaucherel, and made available today on arXiv a new paper entitled Image information content characterization and classification by physical complexity. In the paper we present a method for estimating the complexity of an image based on the concept of Bennett’s logical depth. Unlike the application of the concept of algorithmic complexity by itself, the addition of the concept of logical depth results in a characterization of objects by organizational (physical) complexity. We use this measure to classify images by their information content. The method provides a means for evaluating and classifying objects by way of their visual representations.
The method described in the paper ranks images based on their decompression times and the classification corresponds to the intuitive ranking resulting from a visual inspection, with things like microprocessors, human faces, cities, engines and fractals figuring at the top as the most complex objects; and random-looking images, which ranked high by algorithmic complexity, were ranked low according to the logical depth expectation, classified next to trivial images such as the uniformly colored, indicating the characteristic feature of the measure of logical depth. A gradation of different complexities were found in the groups between, gradually increasing in complexity from bottom to top.
Along the paper we show that:
The paper is available here.