It is very difficult to detect the small target over water in the marine environment because of the complexity of the water movement and the physical field produced by the water interacted with environment. So we mainly studied foreground segmentation for small target in visual image based on Markov Random Fields (MRF). A foreground segmentation method was proposed based on kernel function and MRF. In this method, the special and temporal correlation is used in the kernel function between neighbor pixels. We took the probability distribution of the kernel function method as the observed value of the MRF and get the energy function of the MRF. Results show that our method is applicable to low speed small target segmentation in sea image.