REAL-TIME IMAGE PROCESSING AND IMAGE UNDERSTANDING USING A COMBINED NEURAL NETWORK/EXPERT SYSTEM

THIS PROPOSAL IS MOTIVATED BY THE COMPLEMENTARY CAPABILITIES OF EXPERT SYSTEM AND NEURAL NETWORK (NN) TECHNOLOGIES, AND BY THE POTENTIAL FOR A COMBINED SYSTEM TO PERFORM MORE EFFECTIVELY THAN EITHER ALONE. THE LOCAL, BOTTOM-UP APPROACH TO VISION CANNOT DEAL WITH THE NOISE AND THE ON-GOING PRODUCTION OF NOVELTY IN REAL VISUAL SCENES, WHILE EXPERT SYSTEM APPROACHES ARE TOO BRITTLE AND SLOW TO DO REAL-TIME IMAGE PROCESSING. WE BELIEVE THE COMBINED NN/EXPERT SYSTEM ARCHITECTURE IN THIS PROPOSAL WILL OVERCOME MANY OF THE PROBLEMS WITH PREVIOUS APPROACHES. THE PROPOSED PROJECT HAS 2 PRIMARY GOALS: 1) TO CONSTRUCT A COMPUTER APPLICATION THAT COMBINES A NEURAL NETWORK (NN) COMPONENT AND AN EXPERT SYSTEM COMPONENT IN A WAY THAT ENHANCES THE PERFORMANCE OF EACH; 2) USE THE SYSTEM TO CREATE FASTER MORE FLEXIBLE SOLUTIONS TO EXPERT LEVEL PROBLEMS IN VISION.