Impulse noise removal from color images: An approach using SVM classification based fuzzy filter
Amarjit Roy, Joyeeta Singha, Rabul Hussain Laskar · 2017
This paper proposes support vector machine (SVM) based fuzzy filter for restoration of corrupted images from impulse noise in color image domain. In this proposed filter, filtering technique is performed using channel by channel operation on the corrupted pixel (R, G, and B channel separately). In this work, the system has been trained with the optimal feature set. During the testing phase, pixels within test image are classified using knowledge achieved during the training phase into noisy and non-noisy. If the pixel is classified to be noisy, fuzzy filtering operation is done on the noisy pixels and otherwise, it will be kept as it is. The experiment has been performed on a large set of images and the proposed filter is compared with some of the baseline system likely MSVMAF, and BDND filter. It has been observed that the proposed filter not only tries to improve the image details but also requires less computational complexity while preserving images from its corrupted version.