 |
    |
Memetic Optimization of Video Chain Designs
Walid Ali1 and Alexander Topchy2
1Philips Research Laboratories, 345 Scarborough Rd, Briarcliff Manor, NY, 10510, USA
walid.ali@philips.com
2Department of Computer Science and Engineering, Michigan State University, East Lansing, MI, 48824, USA
topchyal@cse.msu.edu
Abstract. Improving image quality is the backbone of highly competitive display industry. Contemporary video processing system design is a challenging optimization problem. Generally, several video algorithms must be sequentially applied to real-time video data. Overall image quality depends on the nonlinear interactions between multiple design parameters: variable settings for each module (algorithm), the amount of data being transferred in the video processing chain as well as the order of the cascading modules. Unfortunately, no systematic techniques are currently available to configure the video chain without lengthy trial and error process. We propose a rapid and reliable method for optimization of composite video processing systems based on genetic algorithm coupled with local search heuristics. Video system configuration is evolved toward the best image quality, driven by an objective video quality metric. We analyze several local search approaches, including hill-climbing, simplex and estimation-of-distribution algorithms. Experimental study demonstrates superior performance of memetic strategies over the conventional genetic algorithm. We obtain novel and practical video chain solutions that are typically not attainable by regular design process.
LNCS 3103, p. 869 ff.
lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2004