Mean Square Error Reduction Using Genetic Algorithm

Abstract

This work is devoted to design an d analysis o f several aspects of image com pression, specially quantization and coding. Here we utilize a coding technique, which not only preserves some o f the statistical characteristics of the block during quantization, but also giving a fixed bit rate with very easy hardware implementation. Though the approach is very simple but due to limited quantization levels does not perform equally well in every region, resulting in ragged edges an d introduced noise at edges. This work stands in contrast to the above algorithm and implemented an idea based on mean square error criteria, which reduces the above artifacts to a great extent. A natural processing concept called genetic algorithm: a stochastic global search and optimization approach that mimic the metaphor o f natural biological evolution, has been applied to fin d out the optimal solution in a multimodal search space. The multilevel quantization is modeled as optimization problem an d an attempt is made for selecting the better thresholds using GA incorder to reduce mean square error. The simulations results indicate that both the computational complexity and the reconstructed image quality achieved have been improved as a outcome o f this work.