Non-linear filtering of compression blocking effects in reconstructed images

The increasing use of data compression by space mission experiments poses the question of quality of the images obtained after the compression-decompression process. Indeed, working on an Image Compression Module (ICM), Using Discrete Cosine Transform (DCT), with 8*8 pixel-sized sub-images (each pixel being coded on eight bits), one can find blocking effects on their boundaries. Avril and Nguyen (1992, thereafter ANG 1992), have shown that One Neighbour Accounting Filters, used after image reconstruction without modifying the coding method , provide the best and fastest correction as far as linear filtering is concerned. We present here a non-linear method, also used after image reconstruction, but working on spatial frequencies. It allows us to segregate, in the Fourier space, the signal from the defect, and then to remove it through applying a filter adapted to the frequency spectrum of each spoiled image. Employing the reverse Fourier transform, we then retrieve the corrected image. The efficiency of this new method was tested by three different means:- when Fourier filtering is applied to a reference set of aerial photographs of the Earth, blocking effects are quite indistinguishable by human vision, even when zooming on the images, which was not the case with ONAF;- the improvement of the Root Mean Square (RMS) Error, calculated between the filtered and original images, is at least three times greater than the one obtained with ONAF;- the reconstruction of a three-dimensional view of a landscape, thanks to two stereoscopic images having undergone a compression-decompression process with an algorithm using DCT and a compression rate of about 10, is possible only after Fourier filtering has been applied.The quite good preliminary results of the application of Fourier filtering to the Clementine images of the Moon are also represented.