A dominant-noise discrimination system for images corrupted by content-independent noises without a priori references

Abstract

A dominant-noise identification system used on images corrupted by content-independent noises without a priori information has been developed. As representative examples of content-independent noises additive white Gaussian noise (AWGN) and random impulse noise (RIN) are used. The system uses two noise detectors to estimate the intensity component of either AWGN or RIN in an image. For this purpose an impulse noise detector was developed based on directional edges [1] and a Gaussian noise intensity detector developed in [2] was implemented. Then in the second phase, lookup tables that are generated beforehand using the structural similarity index (SSIM) proposed in [3] are used to convert the detected noise intensities onto a common scale to compare the distortion caused by different noise types and identify which one is dominant. Simulations were conducted on over 300 images and results show that for images corrupted with moderate noise intensities the system's overall success rate for identifying the dominant noise is 7 out of 10 times.