A New Unified Impulse Noise Removal Algorithm Using a New Reference Sequence-to-Sequence Similarity Detector

Existing detectors are based on using the similarity of pixels or blocks to locate noisy pixels. Alternatively, this paper presents an approach that investigates the similarity property among sequences of pixels to establish a new reference sequence-to-sequence similarity (RSSS) impulse noise detector. Then, to harness the advantages of this new RSSS detector in high detection accuracy, a new unified image denoising algorithm (RSSS-I) is introduced to remove different types of impulse noise. In this approach, the RSSS detector locates the impulse noise, and then, three different median filters remove the detected impulse noise in a cascade framework. In this framework, an existing weighted median filter is utilized as the domain filter, and two new directional mean and “extreme” median filters are applied as the post-filter. Experimental results show the benefits of this cascade framework in improving the performance. Comparison results demonstrate that the RSSS-I outperforms several existing methods for the ability to accurately locate the positions of noise, retain edge information, inhibit residual artifacts from occurring, and generating denoised images with better quality.