Robust Statistics Technique for De-noising of the Images Corrupted by Uncorrelated or Correlated Noise
Ashish Dixit · International journal of tomography and simulation · 2014
Images are frequently corrupted with noise during capture, digitization, and transmission from one point to another and again during retrieval from storage media. When the images are corrupted with high noise level at the formation stage itself e.g. in the case of thermal imaging or infra red imaging based night vision etc., the noise obstructs the image recognition as it gives a grainy, snowy or textured appearance to the image. So there exists a need for efficient image de-noising method with minimal artifacts in the original image. The dark images, e.g. night shoots, are highly noisy and have very low dynamic range of brightness. These images need to be carefully tackled by the image processing algorithm to produce acceptable visual quality, e.g. surveillance applications. Also the requirement for effective and efficient image de-noising methods has increased with an enormous and trouble free production of digital images and movies by amateurs to professionals. De-noising is also necessary as a pre-processing step in image compression, image segmentation, recognition etc. De-noising has been an important and widely studied problem in image processing and computer vision. Basically, the image de-noising methods are divided into two types: local and non-local. The methods that only exploit the spatial redundancy in local neighborhoods are referred as Local methods. On the other hand, the methods that estimate pixel intensity based on information from the whole image and thereby exploiting the presence of similar patterns and features in an image are referred as Non-Local methods. A non local method called as Non-Local Means [6] estimates noise-free pixel intensity as a weighted average of all pixel intensities in the image, and the weights are proportional to the similarity between the local neighbourhood of the pixel being processed and the local neighbourhoods of surrounding pixels. The proposed algorithm explores a new approach of Non-Local Means estimation by using robust estimation in place of the more usual exponential function.