High performance impulse detector using anisotropic diffusion

Since switching schemes for removing impulse noise detect the positions of pixels corrupted by impulse noise and selectively only filter out corrupted pixels, the restored image quality can be significantly improved compared with conventional median filters. However, when some impulses cannot be detected with impulse detectors, they remain behind, and may adversely degrade subjective image quality. To solve this problem, we propose a high-performance impulse detector based on anisotropic diffusion that smoothes out the details of images while preserving impulses, so that no impulses are left undetected. The experimental results revealed that the proposed approach could effectively improving objective and subjective image quality, especially when images were seriously corrupted with constant or randomly valued impulses.