ΑΛΓΟΡΙΘΜΟΙ ΚΑΙ ΑΡΧΙΤΕΚΤΟΝΙΚΕΣ VLSI ΓΙΑ ΣΥΜΠΙΕΣΗ ΨΗΦΙΑΚΩΝ ΕΙΚΟΝΩΝ

Abstract

IN THIS DISSERTATION WE FOCUS ON HYBRID PREDICTIVE-TRANSFORM ALGORITHMS FOR DIGITAL IMAGE COMPRESSION, FOR WHICH WE PRESENT IMPROVEMENTS AS WELL AS VLSI ARCHITECTURES FOR THEIR REAL-TIME IMPLEMENTATION. FIRST, WE INTRODUCE A ROTATION BASED ALGORITHM FOR THE DISCRETE COSINE TRANSFORM (DCT), WHICH REALIZED WITH THE USE OF A NOVEL DISTRIBUTED ARITHMETIC ROTATION CIRCUIT, LEADS TO TOTALROM REQUIREMENTS LESS THAN 10% OF THOSE OF OTHER KNOWN DCT IMPLEMENTATIONS,WHILE OPERATING AT THE SAME SPEED WITH THEM. NEXT, WE PRESENT A VERY EFFICIENT PARALLEL ARCHITECTURE DERIVED UNDER THE FRAMEWORK OF TIME- RECURSIVE COMPUTATION, FOR THE REAL-TIME IMPLEMENTATION OF THE MODULATED LAPPED TRANSFORM (MLT), WHICH IS USED FOR ALLEVIATING THE BLOCKING EFFECT APPEARING IN VERY LOW BIT RATE CODING APPLICATIONS. THEN, WE PRESENT FLEXIBLE SYSTOLIC ARCHITECTURES IMPLEMENTING FULL-SEARCH BLOCK-MATCHING FOR MOTION AND DISPARITY ESTIMATION, WHICH ARE VERY IMPORTANT IN PREDICTIVE CODING TECHNIQUES. NEXT, WE INTRODUCE FAST MRF (MARKOV RANDOM FIELD) ALGORITHMS FOR EDGE DETECTION, EDGE-PRESERVING NON-LINEAR IMAGE SMOOTHING AND OPTICAL FLOW ESTIMATION, AND SHOW THAT THEY CAN BE USED TO DRAMATICALLY IMPROVE THE PERFORMANCE OF HYBRID PREDICTIVE-TRANSFORM CODING ALGORITHMS. FINALLY, WE INTRODUCE VLSI ARCHITECTURESFOR THOSE MRF ALGORITHMS, ABLE TO FORM THE PROCESSING CORE OF AN MRF BASED APPLICATION SPECIFIC IMAGE PROCESSING SYSTEM.