Image Enhancement Based on A New Genetic Algorithm

The nonlinear transform of gray level is an efficient method in the field of image enhancement. But in the previous methods, the high computational complexity and the poor robustness are the common disadvantages. Thus it is very significant to find intelligent algorithms on adaptive enhancement of image.Firstly, this paper proposes a novel genetic algorithm(GA), which can not only keep the population diversity but also has quicker convergence speed. Our idea is that since the search ability of crossover in binary coding is better than that of decimal coding, it is reasonable that GA employs binary coding with several mutation bits to improve the performance. As the number of mutation bits increases, however, GA may become random search. To overcome the above shortcomings, our approach obeys the rules as follows.The number of adaptive mutation bits in individual i, M i=(int)N×f max-f if max-f min, where N is a constant, f max and f min are the maximum and minimum fitness values of the population respectively, f i is the fitness value of individual i. And f max-f min is the range of fitness value of solutions in the population. The value of term, F=f max-f if max-f min, is a yardstick for presenting the degree of goodness of individual i in the population. The value of F is normalized to the range 0 0—1 0. The smaller F is, the better the fitness value of individuals is and vice versa. By using M i, the number of mutation bits of individuals is varied adaptively depending on the fitness values of the solutions, i.e., the high fitness solutions are protected from disruption by undergoing mutation with fewer bits while the low fitness value solutions are modified by more mutation bits to prevent GA from getting stuck at a local optimum. Secondly, image enhancement is done. As for humans' visual sense, there are three states for most of the gray-level images. Accordingly, four functions are used to transform gray level of images. To simulate the four kinds of transform functions stated above, Tubbs proposed a normalized incomplete Beta function B(α,β).With the different values of α and β, the four functions can be simulated. We employ our proposed GA to optimize α and β adaptively according to the quality of image. Our experiments show that this method is practical and efficient.