On modeling the distribution of chest X-ray images

Summary form only given. One of the determining factors in parametric modeling of a stationary image source is its marginal probability distribution. There have been several different assumptions about this distribution, based on either histogram measurement with an ergodicity assumption or the physics of the image-generating process. Gaussian, Rayleigh, exponential, and some other distributions have been reported to model the source. It is shown that the probability density function of the differential image can be very well modeled as a generalized Gaussian distribution. A Peano-type differential operation, which has been shown to be the optimal scanning method and essentially achieves the entropy of the image asymptotically, has been implemented. The Kolmogorov-Smirnov test for goodness of fit has been used for 20 normal chest X-ray images. On the basis of the test results a first-order generalized Gaussian autoregressive model for the image source has been proposed and its properties and applications studied