The look for proficient image denoising methods still is a substantial task, at the intersection of practical analysis and measurements. Disregarding the refinement of the as of late proposed methods, most algorithms have not yet achieved an alluring level of relevance. All demonstrate an exceptional performance when the image model relates to the algorithm presumptions, however unable to do as a general and make ancient rarities or evacuate image fine structures. The principle center of this paper is, to start with, to characterize a general mathematical and test methodology to think about and arrange established image denoising algorithms, second, to propose an algorithm (Non Local Means) tending to the safeguarding of structure in a digital image. The mathematical analysis depends on the analysis of the "method noise", characterized as the contrast between a digital image and its denoised form. The NL-implies algorithm is turned out to be asymptotically ideal under a non specific factual image model. The denoising execution of every considered method are looked at in four ways; mathematical: asymptotic request of extent of the method noise under consistency presumptions; perceptual-mathematical: the algorithms curios and their clarification as an infringement of the image model; quantitative trial: by tables of L2 separations of the denoised variant to the first image.