Abstract: Anisotropic diffusion filter, being an edge preserving non-linear image restoration method, has been of wide acceptance and of intrinsic interest among the researchers in image processing domain. But judicious choice of its operational parameters is mandatory to ensure the optimum performance. Its operational parameters include number of iterations, threshold of gradient modulus etc. These operational parameters are usually selected through trial and error, through qualitative inspection of restored image. The noise retains even after the restoration and the image is likely to get blurred if the number of iterations is not wisely chosen. The selection of threshold of gradient modulus is not image driven and usually blindly chosen. Moreover, in anisotropic diffusion the diffusion coefficient is computed from the local gradient. Even the noise pixels may exhibit high local gradients in heavily corrupted images. The fine details in the image may be lost during the iterative smoothing unless the operational parameters are carefully selected. The main objective is to propose an adaptive image driven estimation of threshold of gradient magnitude, rather than setting it manually. The performance of the modified anisotropic diffusion will be compared with conventional P-M model in terms of the ability to preserve edges during restoration.

Keywords: Image Quality measures include Edge Content Ratio(ECR),Pratt’s Figure of Merit(PFOM) and Percentage Reduction in the Standered deviation of Noise(PRNSD).