Speed up of Reindexing in Adaptive Particle Swarm Optimization

Abstract: Palette re-ordering is a class of pre-processing method with the objective to manipulate the palette index such that the adjacent symbols are assigned close indices in the symbol space, thus enhancing the compressibility of the image with many lossless compressors. Finding an exact reordered palette would certainly be exhaustive and computationally complex. A solution to this NP hard problem is presented by using an Adaptive Particle Swarm Optimization (APSO) to achieve fast global convergence by maximizing the co-occurrences. A new algorithm with improved inertia factor is presented here to accelerate the convergence speed of the reindexing scheme. In this algorithm, the key parameter inertia weight is formulated as a factor of gradient based rate of particle convergence. Experimental results assert that the proposed modification helps in improving APSO performance in terms of solution quality and convergence to global optima.

Key words: Reindexing, palette-indexed image, Cross Entropy (CE), rate of particle convergence (k), improved Inertia Weight Adaptive Particle Swarm Optimization (IWAPSO).

Received April 3, 2013; accepted November10, 2014