A New Denosing Method Based on Local Similarity for Removing Impulse Noise

Due to the randomness and complexity of impulse noise, it is a difficult task to remove impulse noise from damaged natural images. This paper proposes a two-stage denoising method in order to effectively remove low-to-medium-density impulse noise. In the noise detection stage, the LCI impulse noise detector was studied and analyzed, and then the clean edge pixels were screened out from the candidate noise by introducing the edge pixel detector, so as to achieve the purpose of reducing the false detection rate of noise. In the second stage of noise removal, a filter based on LCI weight is designed. The filter takes all pixels in the neighborhood of the center pixel as the reference object, but the weight of each pixel in filtering is related to its own local similarity, which can reduce the interference of pixels suspected of noise to the filtered pixels. By testing on common test images, it can be found that the improved noise detector can accurately and quickly detect the impulse noise. In terms of image quality restoration performance, the filtering effect of the proposed filter is better than or comparable to other noteworthy methods in objective indicators. In addition, the denoising method has great advantages in terms of running time and model complexity.