A fuzzy logic approach for the fast approximate computation of image transforms from block JPEG DCT coefficients

Computationally efficient implementation of different image processing and analysis algorithms is crucial in many modern real world applications. Several modern visual analytics applications involve processing large amounts of compressed image/video data on restricted hardware resources, in a numerically efficient manner. In this context, the formulation of some image analysis/processing algorithms in the compressed image/video domain can bring significant benefits in the computational complexity reduction. An important step to image analysis, feature extraction, based on unitary image transforms is addressed in this paper, in a general form. Starting from the fact that in most popular video compression the intraframes are block discrete cosine transform (DCT) encoded, we present and propose a solution to computing any unitary image transform at pixels block level, starting from the DCT coefficients of the block. Since approximate forms of the transforms are often satisfactory, we propose a fuzzy logic system based approach to optimize the computational complexity with minimal loss of image information. The method is exemplified for the case of the Walsh transform, which shows that even the fastest implementation provides (by reconstruction) images of good quality.