Genetic Algorithms And Active Contour Method The Picture Text Extraction

Come along with the information age,people pay more and more attention to information on the Internet,and make use of them.A lot of information is digitalized by the multimedia information what basically includes the text, picture, audio and video. So,it becomes a hot topic to classification of the multimedia information.People usually search for the information by keywords.The multimedia information started to appear, people added keywords for these multimedia information by themselves, but we find that there are many multimedia information (such as pictures, videos), what contain the keywords describing themselves, thus, how to extract characters in images becomes a topic what is paid more and more attention by people.First of all,this thesis discusses several classic methods of the character extraction,and we get a conclusion what these methods rely on the characteristic of character. Actually, we can know character extraction is a problem of the image segmentation. And then, it researches the popular active contour method. We realize character segmetation that color of the character is close to background's and the character has the deep groove, using Snake model of based on the level set function. And then, we use genetic algorithms to solve the problem that it hardly made sure to the paramenters of variational Snake model. Finally, the above method are applied in the character extraction, and we do the related experiments in MATLAB.The main contents of this thesis are as follows:Firstly,it introduces that we use the method of character localization in this paper: active contour method (variational Snake model).This model is based on the level set function of Snake model,and is a energy function what is composed of the internal energy and external energy function. The internal energy function makes the level set function keep close to SDF(signed distance function), the external energy function drives the zero level set curve of the level set function to move to the target boundary.When the total of the energy function is minimum, it shows that the curve has arrived the target boundary.Secondly, the parameters of the variational Snake model are difficult to be determined, so we use the genetic algorithm to solve the parameters' problem. The role of genetic algorithm searches the parameter values in the range of the parameters.Firstly, we get multigroup parameter values(population) by random generation, and then calculate the fitness value of each group parameters(individual), and then select the individual according to fitness value,crossover and mutation operation for these individuals.Finally,we will get a rational parameters group. The fitness depends on the objective function, and the objective function is the energy function of variational Snake model.Finally, it is applied in the experiment that the genetic algorithm combines with the variational Snake model. Firstly, the picture is pre-processed. Secondly, we initialize the level set function, and create the initial population. Finally, we will get the final result by calculating the objective function values, calculation of fitness values, genetic operation.