Bispectrum-Based Detection Homogeneous Areas in the Digital Images

A novel bispectrum-based strategy applied to digital image processing is suggested for solving the problems of detection the homogeneous areas in digital images. Novel classification features evaluated in the form of bispectrum magnitude estimates and computed for the certain image areas are suggested and studied. Decision about homogeneous area has been made by comparative analysis of the bimagnitude peak values. Small biamplitude values indicate homogeneous area and large biamplitude values indicate non- homogeneous area contained in the image. Results of computer simulations performed for grayscale images are represented and discussed. Performance of detection the homogeneous areas has been examined both for various noiseless and contaminated by noise test images. The biamplitude computations have been performed under influence of four various noise types. It has been examined the following noises: Gaussian, Poisson, salt & pepper, and speckle. According to the experimental results, proposed technique provides high accuracy for detecting homogeneous areas in the noiseless images. But in the presence of noise, the proposed technique is less effective.