An efficient pipelined architecture for multilevel wavelet based image denoising
Jyeshtharaj Bhalchandra Joshi, Nisseem Nabar, Rohan Adyanthaya, P. Batra · 2006
A pipelined architecture for denoising images based on the statistical modeling of wavelet coefficients is presented. The basic algorithm calls for approximating the noisy wavelet coefficients based on observing their local neighbourhood. Certain modifications here, based on a previous work have highlighted an aspect of hardware optimization. The wavelet module and the neighbourhood observation module have been designed in a pipelined manner so as to address real time applications along with an attempt towards low embedded memory applications. The architecture has regular data flow and is adaptable to arbitrary image sizes. The wavelet used is the Daubechies' 9/7 biorthogonal wavelet. This paper highlights the advantages of using a higher level of the wavelet transform with the aspect of hardware optimization and better denoising. A comparative study along with a deeper exploration of PSNR results for different test images have been presented along with the hardware results.