AN EVOLUTIONARY BASED IMAGE GENERATOR

Abstract

This paper presents a fast algorithm that utilizes evolutionary algorithm for patch based imagegeneration. The algorithm attempts to find the most appropriate collection of source square imagepatches that can generate altogether a given colored/gray target image. Each patch holds completeinformation for both luminance and chromatic components from the source image to be used togenerate a specified target patch. The difficulties are occurred when chromatic and luminanceinformation is to be transferred to a blank canvas to generate a specified target image. The proposedEvolutionary Algorithm (EA) is used to overcome these difficulties and attempts to find anacceptable and perceived solution to this problem by the search for best patch luminance matchinguntil a specified number of generations is met. Also, this algorithm modifies the traditional uniformcrossover to a multi-sexual recombination operator which inherits vital genes among all individualto one super generated offspring. All other offspring are generated by mutating that super offspring.The experimental results have demonstrated that good results can be obtained with reasonableconvergence speed. Also, results show that when the patch size is increasing, the mean square erroris decreasing.