中文摘要：

为了利用侧扫声呐进行水下目标自动探测和识别，首先必须将声呐图像分为目标高亮区、海底混响区和目标阴影区．由于声呐图像有强背景噪声，传统的图像分割方法显得无能为力，故采用基于MRF场的图像分割方法来准确地分割．根据侧扫声呐目标的成像特点，建立了分割的约束条件；利用阴影与目标的灰度均值比很小这一特点进行初始分割，然后根据分割后目标与阴影的宽度差来剔除虚假目标，由初始分割的结果求得MRF模型初始参数，再采用迭代条件估计得到最终的模型参数和准确的分割结果．由于考虑了相邻像素间的依赖关系，具有抗噪性强、分割效果好的优点，从理论上说是合理的．实测数据分析也证明了这种算法的优越性．

英文摘要：

Side-scan sonar image （SSI） must be segmented into regions of shadow, sea-bottom-reverberation, and object-highlight before underwater object can automatically be detected and recognized. Because strong background noises exist, traditional algorithms of image segmenting are useless. The algorithm based on Markov random field model is introduced. The segmentation can be constrained by the aprior information, according to the characteristics of object on the SSI. Furthermore, it is highlight intensity in an object area and lowlight intensity in a shadow area, so the ratio of shadow intensity to object intensity is very small. The SSI can be initially segmented by the three aprior information. After the initial segmentation has been completed, a false objects can be detected through the characteristic that the difference between the widths of object and shadow is close to one. And then, an MRF model parameter can be solved with the least square, and an noise parameter can be calculated with the maximum likelihood approach. Fi- nally, the segmentation can be accomplished with the ICE method. The MRF model provides a reliable method for obtaining this underlying label field through incorporating pixel dependencies into the segmentation model. This is rational and robust. It has few influences when strong speckle noise exists. This fine result is obtained through the real SSI.