Abstract

Digital images are often corrupted by impulse noise during acquisition or transmission due to the camera sensors, faulty memory locations or environment. The efficiency of noise reduction techniques in digital images strongly depends upon the successful detection of the noisy pixels. The pixels corrupted with salt and pepper noise (SPN) have the value of 0 or 255 in the grey scale digital image that cannot always be true because some non noisy pixels with the value of 0 or 255 may contain important information of the image hence requires the distinction of such precious pixels and noisy pixels. There are numerous filters available in the literature those presume pixels noisy having intensity minimum or maximum, and behave poorly with the increase in levels of noise. A novel quartile based method for the convolution matrix of 3x3 or 5x5 is proposed to label pixels noisy or noiseless having the pixel value of 0 or 255 successfully and have been found efficient subjectively and objectively when compared to standard techniques available in the literature on the basis of parameter for evaluation like PSNR, MSE, SSIM while preserves the sensitive information of the digital image. The proposed framework demarcates pixels based on the local morphology of neighboring immediate quartile pixels. The proposed method only processes noisy pixel while leaves non noisy pixels unchanged.

Authors

Imran Qadir, V. Devendran
Lovely Professional University, India

Keywords

Salt and Pepper Noise, Linear Cellular Automata, Mean Filter, Median Filter, SPN Identification