Sonar image enhancement of digraph and Gaussian mixture model in complex contourlet domain

A sonar image enhancement algorithm based on the directed probability graph of Complex Contourlet Transform(CCT) and Gaussian mixture model is proposed.Using complex Contourlet analysis to extract the weak feature information of each direction of the sonar image in each scale;In order to establish the relationship between the feature information,We consider that the state of subband coefficients between adjacent scales of the complex Contourlet domain has Markov property,and the state of sub-node coefficients depends on the state of parent node coefficients,and constructs a directed probability graph model to reflect this continuity of complex coefficients;Within the scale,we build a Gaussian mixture model to establish the connection of characteristic information in the same scale,and use a two-state Gaussian mixture model to characterize the non-Gaussian edge distribution of subband coefficients.Finally,the Expectation Maximization(EM) algorithm is used to train the model parameters,estimate the coefficients of the enhanced image so that it can realize the sonar image enhancement.The experimental results show that compared with the wavelet domain Hidden Markov Tree(HMT) algorithm and the Contour let domain HMT algorithm,the Peak Signal-to-Noise Ratio(PSNR) of the proposed algorithm increases by more than 4 dB,and the Structural SIMilarity(SSIM) index increases by 0.3;The algorithm in this paper can not only suppress the strong noise of the sonar image,but also retain the weak feature information such as the edge and texture of the image.