Abstract: The loss of information on image details was often found in image denoising process if using the conventionally typical method of impulse noise filtering, which resulted in blurred images. Based on local similarity analysis and neighborhood noise evaluation, a new image denoising algorithm is proposed to analyze the local similarities between all pixels in an image so as to determine the outline and noise of an image. Then, the noises are detected through neighborhood impulse noise evaluation so as to enable the algorithm to just process noise pixels with the pixels of image outlines kept unchanged. In this way, the accuracy of noise detection can be improved more efficiently with image details well preserved. Experimental results showed that the new algorithm outperforms other prior-art methods in suppressing impulse noise and detail preservation, thus offering a new filter applicable to image processing.