The target of this paper is to classify the seafloor automatically based on the sonar images processing in 2D. In order to analyze the seafloor effectively based on the sonar images, we must improve the image quality or gain the region which we are interested in, mainly aimed at speckle reduction, contrast enhancement. Meanwhile, the information from shadow and echo is very important for our further analysis, the detection of each object located on the sea-bottom and its classification are generally based on the extraction and the identification of its associated shadow. Thus, before any classification step, one must segment the sonar image in terms of shadow areas and sea-bottom reverberation areas. Recently, Markov random field (MRF) is widely used in fields of computer vision and digital image processing, which expresses the spatial correlation of the image pixels sufficiently. In this paper, the basic theory about MRF is introduced and some improvement is presented according to the feature of sonar images. We analyze the main factors influenced the sonar images in the process of generation. Furthermore, according to the analysis about the main problem, we propose a scheme to extract the texture feature of sonar images using Gabor filters with optimal parameters, which can solve the directional problem perfectly. The change of the slant angle along the swath or the perpendicular position of the tow fish will result in the variety of the size of the object, as well as the variety of the texture character of the seafloor. Taking the scale or resolution factor into consideration, we propose two more efficient classification algorithms. One is an information fusion method, which combines both Gabor filters and fuzzy fractal dimension, to extract features and to classify seabed physiognomy. Another applies Gabor wavelet to extract the features of sonar images, which combine the advantages of both the Gabor filter and traditional wavelet function.