Image Impulse Noise Removal Using a Hybrid System Based on Self Organizing Map Neural Networks And Median Filters

Digital image restoration has become important for many image applications. Therefore, Image Noise removal is an essential issue in an image processing fields. In this paper, we presented a hybrid system, based on Self Organizing Maps neural networks (SOM NN) and Median filter (MF), to eliminate Random Impulsive Noises (RIN) from grayscaled digital images. In our system we applied two main processes (features extraction process, detection process): in the features extraction process, feature vectors of two or three features, (central pixel value, standard deviation value) or (pixel value, standard deviation, Mean Difference value) respectively, were extracted. In the detection process, the self-organizing maps (SOM NN) are used as an impulse noise detector. This SOM NN module is trained using competitive learning algorithms. Then the detected corrupted pixel is modified using Median filter (MF) algorithm; otherwise, it is left unchanged. Finally, the results of this study are compared with the previous traditional and state-of-the-art methods results that applied on the same database. Our results are significantly outperforms other traditional methods and were comparable with the state-of-the-art methods results in terms of reconstruction quality and that are comparable to the FL using the MSE and PSNR measurements.