Multiscale noise-adaptive homomorphic filtering based speckle denoising in laser speckle imaging

Abstract

Speckle pattern forms as a result of constructive and destructive interference due to the phase difference involved in multiple backscattered photons. In coherent optical probing, speckle has a dual role as an insidious form of noise as well as a source of information for sample microstructure. This paper focuses on designing of a multiscale noise adaptive homomorphic filter (MNAHF) for speckle noise reduction (keeping the signal-carrying speckle) in laser speckle imaging (LSI). The operation has been performed based on the calculation of local statistics within a sliding window region over a successive number of time frames. It has been shown that MNAHF outperforms other speckle denoising filters from the point of image quality assessment and better visualization. It exhibits lower mean square error and Minkowski metric values with better image quality, intact structural similarity and superior signal to noise ratio relative to others. Therefore, over a wider spectrum, this work brings out the scope of speckle noise reduction prior to further large scale processing for automated segmentation, texture analysis and image based classification techniques in clinical study of laser speckle images.