Modified homomorphic wavelet based despeckling of medical ultrasound images

Speckle noise suppression is a prerequisite task in order to maintain the diagnostic potential of ultrasound imaging. Among various despeckling methods, there exists a class which transforms the multiplicative speckle noise to the additive through a logarithmic transformation. In most of such studies, it is assumed that the samples of the multiplicative noise are mutually uncorrelated and they obey the Gaussian distribution. Present studies show that this assumption is oversimplified and it results in inadequate performance of speckle suppression. We introduce an adaptive preprocessing filter which de-correlates the samples of speckle noise and approximates its behavior to that of white Gaussian noise. The study also evaluates the performance of homomorphic wavelet despeckling (HWDS) with this adaptive preprocessing as the initial stage and demonstrates that the proposed adaptive preprocessing stage significantly improves the performance of HWDS both qualitatively and quantitatively.