Relative Entropy of Grayscale Images Degraded by Bit-Plane Quantization and Random Noise

摘要

This paper presents an experimental study of the relative entropy of grayscale images degraded by bit-quantization and additive random noise. Bit-plane quantization is produced by removing any number of selected bit-planes from a given grayscale image, whereas noise is modeled additively using impulse, uniform, Gaussian, Poisson, and Cauchy probability density functions. A comparison of the behavior of relative entropy with other image measures such as the normalized mean square error, the signal-to-noise ratio, and the correlation coefficient is also given for different levels of bit-plane quantization and parameter values of the aforementioned noise probability density functions. An illustrative example is provided together with characteristic graphs that demonstrate quantitatively the overall discrimination capability performed by the relative entropy measure.