ECE 901

OPERATION HAMLET II

Due: Monday, March 30 Rm 3537 before 4:00pm.

Part 2

"Iterative insolence"

Situation

An important transmission from one of our field agents was distorted during transmission. Two versions arrived. The are quite different. Intelligence NSA has found the frequency response derived for the from the distortion of the first pixel in each image was transmitted as loadable Matlab variable: hzf.mat, hsf.mat At Matlab command line type 'load hzf.mat' to load variable 'hzf'. In addition, the noise variance for each was found to be: s_s² = 2.0 s_z² = 4.0. Image Data zdeg.pgm sdeg.pgm Image Variances s_zorig² = 3.019x10³. s_sorig² = 2.716x10³.

Mission

1.) Restore the images using a regularized iterative solution approach in the DFT domain. (Assume spatially invariant distortion.)

2.) Restore each image using 2 different order of magnitude estimates, the optimal value of the regularization parameter as determined from the SNR relationship in Katsaggelos and a subjective best.

3.) Record and plot data fidelity residual, ||g - Hf||, at each iteration.

4.) Stop iteration when, ||f_k+1 - f _k|| / ||f_k|| < 10^-5 (Katsaggelos).

5.) Report PSNR results for OPERATION HAMLET I.

Best results for plane image in each group with C = Laplacian.

For generalized and regularized inverse.

Report results for C = H-I (optional).

Higher the PSNR the better.

Explain results. Is PSNR a reliable quality measure?

PSNR tool : psnr.m

Original Images: plane512.pgm, plane256.pgm

6.) In Hamlet II, both distortions are the same extent. The visually worse distortion

may give better restoration results. Why (not)?

7.) Report all results and activities.

PROJECT DUE: Monday, March 30, 1998 before 4:00pm in Rm. 3537, Agent Jean Stover's Office.