Information-Theoretic Entropy Approaches and their Applications to Texture Analysis
Résumé
The concept of entropy, introduced in this chapter, connects image processing with information theory. The amount of information and the rate of information production, respectively calculated using Shannon entropy and conditional entropy, are two main nonlinear entropy approaches becoming increasingly more popular in image processing. Based on these concepts, two-dimensional approximate entropy, sample entropy, fuzzy entropy, permutation entropy, distribution entropy, and dispersion entropy were developed to estimate the irregularity or uncertainty of images, which has been shown to be related to its texture properties. The estimation of irregularity at different spatial scales (multiscale entropy approaches) was also proposed to assess the complexity of images. In this chapter, the basic concepts of entropy and entropy rate are presented, together with the interpretations of one-dimensional entropy methods. Next, a detailed description of the two-dimensional entropy methods is provided. Then, the advantages and disadvantages of each method are discussed in detail. In addition, some real-world applications of the bidimensional entropy metrics are presented. Finally, the conclusions suggest several lines of future research. Those bidimensional entropy approaches, as well as their multiscale extensions, play prominent roles in the texture analysis of two-dimensional data.