Unsupervised sonar image segmentation method based on Markov random field

Utilizing the characteristics of sonar images, a new unsupervised method is proposed to segment sonar images automatically based on Markov random field(MRF). The research demonstrated that the histogram of sonar images in reverberation area obeys the rule of Gaussian distribution. However, its discrete distribution effect is not beneficial to the automatic segmentation. In this paper, a fast and effective Gaussian pyramid model is used for the preprocessing of sonar image, in an attempt to make the histogram of the bottom reverberation of these images obey Gaussian distribution. On this basis, a model that may automatically determine the number of sonar images classification is proposed. By combining this model with a local energy extremum method, the initialization parameters of the MRF model were estimated to form a fully automated sonar image segmentation model. Finally, the model can be used for segmentation experiments on the data of sonar images, and it is compared with other typical segmentation algorithms, verifying the efficiency and rapidity of the method. ©, 2015, Editorial Board of Journal of HEU. All right reserved.