Advancing Three-Stage Impulse Image Noise Removal Methodology via Neural Networks
Subject Areas : Electronic
1 - استادیار، گروه مهندسی صنایع، واحد سمنان، دانشگاه آزاد اسلامی، سمنان، ایران
Keywords: Impulse Noise, Image Processing, Neural Networks, Salt and Pepper Noise,
Abstract :
Image performance often faces jeopardization due to the presence of undesired noise. Among the various forms of noise contamination, impulse noise remains a significant concern for digital images. Two primary categories of impulse noise are recognized: salt and pepper noise, and random valued noise. Salt and pepper noise perturbs images by causing individual corrupted pixels to adopt either the maximum or minimum gray level value.This study presents a novel approach for the identification and restoration of Salt and Pepper Image Impulse Noises, employing a three-tiered framework utilizing distinct Trained Artificial Neural Networks (ANNs). The noise detection phase addresses two distinct categories of salt and pepper noises: White Range Noises and Black Range Noises. The initial artificial neural network (ANN-I) is dedicated to the identification of white range noises within images afflicted by salt and pepper noise. Subsequently, the second trained artificial neural network (ANN-II) is employed for the restoration of the previously detected white range noises. In the final stage, attention is directed toward detecting remaining black range noises and their subsequent restoration through the utilization of ANN-II, which has been specifically trained to rectify damaged pixels. To evaluate the efficacy of the proposed algorithm, two critical performance metrics are employed: Peak Signal-to-Noise Ratio (PSNR) and Mean Square Error (MSE). These metrics are computed and subsequently compared against those produced by established filters for a variety of test cases.