Hi,
Here is a thought. Suppose the noisy image is im_noisy.jpg. Initially apply a suitable filter or denoising method to denoise it. Let the image be now im_ref.jpg Use this as the reference image and run the LMS algorithm. After a number of iterations, like when the output image becomes a close approximation of the reference image. Make the output image the reference image now and keep running the LMS algorithm till you reach a sufficiently denoised image...