A self-constructing type-2 fuzzy neural network for impulse noise removal in digital images

We propose a self-constructing type-2 fuzzy neural network for impulse noise removal in digital images. The architecture and corresponding parameters of the network are initialized by a SVD-based self-constructing rule generation algorithm, and the initialized parameters are optimized by a hybrid learning algorithm. The trained network can be employed to detect the noisy pixels in the images corrupted by impulse noise. After that, pixels identified as noisy are processed by a median filter whereas the other pixels are retained. Compared with the other approach, experimental results have shown that our approach possesses a better detection capability of impulse noise, resulting in the more effective filtering of impulse noise.