IMPROVED FORWARD-LOOKING INFRARED (IR) IMAGE SEGEMENTATION BY ITERATIVE TECHNIQUES: ALGORITHMS FOR REAL-TIME PROCESSING

ALGORITHMS WILL BE DEVELOPED FOR ENHANCEMENT AND SEGMENTATION OF FLIR IMAGES, WITH A VIEW TOWARD REAL-TIME IMPLEMENTATIONS ON SEQUENCES OF IMAGES ACQUIRED AT VIDEO-FRAME RATES. THE METHODOLOGY PIONEERED BY PERSONNEL ON THIS PROJECT, USING PROBABILISTIC MARKOV RANDOM FIELD MODELS AND COMPUTATIONAL METHODS SUCH AS SIMULATED ANNEALING (STOCHASTIC RELAXATION WITH ANNEALING), WILL BE EXTENDED IN TWO DIRECTIONS: (1) NEW DETERMINISTIC VERSIONS OF THE ALGORITHMS WILL BE APPLIED TO INFRARED IMAGERY, SUBSEQUENT TO CAREFUL MODELING OF ENVIRONMENTAL AND SENSOR EFFECTS ON THE OBSERVED IMAGES; (2) THE MARKOVIAN MODELS WILL BE EXTENDED TO USE TEMPORAL AS WELL AS SPATIAL INFORMMATION FOR IMAGE RESTORATION. FINALLY, THE ALGORITHMS WILL BE DESIGNED AND OPTIMIZED FOR POSSIBLE REAL-TIME IMPLEMENTATION ON A STATE-OF-THE-ART PARALLEL IMAGE PROCESSING SYSTEM. THE SUCCESS IN ATTAINING THE LAST OBJECTIVE WILL DETERMINE THE FEASIBILITY OF PERFORMING TRUE REAL-TIME ENHANCEMENT AND SEGMENTATION OF FLIR IMAGES AS A PRECURSOR TO HIGHER-LEVEL TASKS OF OBJECT DETECTION/RECOGNITION.