This paper presents a new impulse noise filter based on pulse coupled neural networks according to the apparent difference of gray value between noised pixels and the pixels around them. Comparing with the state-of-the-art denoised PCNN filter, the step by step modifying algorithm based on PCNN also, the new PCNN filter suggested in this paper costs less computation and less execution time. At the same time this new PCNN filter has been compared with other nonlinear filters, such as median filter, the stack filter based on omnidirectional structural elements constrains, the Omnidirectional morphology Open-Closing maximum filter (OOCmax) and the Omnidirectional morphology Close-Opening minimum (OCOmin) filter. The results of simulation shows that this algorithm is superior to standard median filter, the state-of-the-art PCNN filter, the maximal, minimal morphological filter with omnidirectional structuring elements, and the optimal stack filter based on omnidirectional structural elements constrains in the aspect of the impulse noise removal. What is more important is that this algorithm can keep the details of images more effectively.