Statistical image modeling using distribution of relative phase in the complex wavelet domain

In this paper, the probability density function of relative phase (fRP) is proposed for modeling natural images in the transform domain. We demonstrate that the fRP fits well with the behaviors of relative phase in the complex directional wavelet subband from different natural images. Moreover, a new image feature based on the fRP is proposed for texture image retrieval application. The fRP based feature yields a higher retrieval accuracy compared to the energy features, the relative phase features and the generalized Gaussian density based features (GGD). In addition to the GGD based features, the fRP phase information is also incorporated to further improve the performance.