Adaptive Impulse Noise Removal Using a Cost Function Based Optimal Partitioning

This paper presents a new impulse noise detection and removal technique based on applying dynamic optimal partitioning (OP) to a set of neighborhoods of a pixel whose noise identity is in question. Using the nature of the impulse noise, and by gathering collaborating information from different directions around it, a pixel is deemed either noisy or normal. If the pixel is classified as noise, then a median-based noise filtering technique, or any other appropriate filtering technique, is applied; otherwise, the pixel is considered normal and left unaltered. The noise detection algorithm uses an effective dynamic optimal partitiong technique that incorporates a noise-based cost function and works for any size of neighborhood without any major algorithmic adjustments. Different cost functions are introduced for the algorithm with simulation results that show the detector's effectiveness in the presence of low to moderate impulse noise levels.