Super Pixel Segmentation with Neuro-Fuzzy Filtering based Complex Impulse Noise Removal for Color Images

Abstract

In image processing, noise filtering or removal is the most essential task to eliminate the corrupted image pixels from a given image and restore the noiseless image. Many filtering techniques with decision mechanisms have been developed to remove the impulse noise from the images. Among those techniques, a super pixel was detected by segmenting the given color image by using mean shift filtering followed by a clustering based on quaternion color distance. The detected super pixel characteristics were analyzed to classify those as noise-free or noisy or single-point impulse ones. After that, those were eliminated based on a Selected Recursive Vector Median Filter (SRVMF) with adaptive window sizes. However, the performance of this technique depends on the orientation of the input image. Therefore in this article, Super Pixel Segmentation (SPS) with Neuro-Fuzzy network based Filtering (NFF) technique is proposed for color images. Initially, the image is segmented to detect the super pixel followed by classification process based on the analysis of super pixel characteristics in order to classify the super pixel as noise-free, noisy and single-point impulse ones. After that, NFF with hybrid learning rule is applied on the detected noisy pixels to remove those from the images with reduced computational complexity and running time. Finally, the experimental results demonstrate that the proposed technique achieves a better denoising effect and performance compared to the other color image denoising techniques.