Abstract
Texture is described in several approaches by 1st and 2nd order statistics which cannot preserve phase information carried by the Fourier spectrum. Besides, these statistics are very sensitive to noise. In this paper, we study features derived from higher order spectra, especially the third order spectrum, namely the bispectrum, known to offer a high noise immunity and to recover Fourier phase information. In this paper, we exploit phase preservation property by using bispectrum phase. We propose wrapped Cauchy distribution to model phase. Wrapped Cauchy parameters are estimated by maximizing the log-likelihood function. Experiments show that the wrapped Cauchy distribution fits our phase information well. Hence, their parameters are used to feed our feature vector in order to classify textures corrupted by Gaussian noise. Classification results using the proposed approach show a good noise immunity compared to a statistical model based on Gabor phase.
Highlights
Texture analysis is an important area of study in computer vision that seeks to find an efficient description of textures
Texture is described in several approaches by 1st and 2nd order statistics which cannot preserve phase information carried by the Fourier spectrum
We study features derived from higher order spectra, especially the third order spectrum, namely the bispectrum, known to offer a high noise immunity and to recover Fourier phase information
Summary
Texture analysis is an important area of study in computer vision that seeks to find an efficient description of textures. Several methods for describing texture have been proposed in literature and they are classified into structural and statistical Most of these methods do not take into consideration noise, this is the reason why we are interested In this paper by features derived from higher order spectra especially the third order spectrum namely the bispectrum which can be classified as statistical. Algorithms based on bispectrum have proved to be able to deal with non gaussian and noisy images in applications such as pattern recognition [1], image restrieval [2], texture classification [3] and biomedicine [4] [5] These algorithms exploit higher order spectra properties. Higher order spectra as in the case of bispectrum, satisfy invariance properties [7]and others important properties such as insensitivity to gaussian noise and phase preservation :.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
