Image Edge Detection Method Based On Ant Colony Algorithm

As a kind of simulated evolutionary algorithm, ant colony algorithm mainly uses the information to choose the direction and then get a better solution through global search. With such advantages as positive feedback, distribution and robustness, this algorithm has been widely applied to network routing optimization, vehicle scheduling, data mining, swarm intelligence and so on.The traditional ant colony algorithm is easy to fall into the local optimal solution and cause stagnation, so to solve these problems, this thesis comes up with an improved ant colony algorithm. This algorithm, through introducing bee colony algorithm idea, divides them into different groups and gives them different path to explore the probability. And at the same time, the algorithm combines initial pheromone optimization and information pheromone updating strategy to improve its defects and find out global optimal solutions. This algorithm uses the eil51 data set to examine the main parameters such as the number of ants, pheromone content and volatile factors, and then the optimal setting range of each parameter are obtained. To tackle such problems as fuzzy pseudo edge and edge continuity which appear easily in the classic methods of image edge detection, the author applies the improved ant colony algorithm into image edge detection and obtains a better edge information confirmed by experimental results.The main work and achievements of this thesis are as follows:(1) Using the classification method of bee colony algorithm, ant colony algorithm divides the ant colony into three different groups, namely the detecting ants, leading ants and following ants, each being responsible for different exploration paths. This classification method can not only save the time of convergence and get a better path solution set, but also fully reflect the characteristics of collaborations of ant colony.(2) For different groups of ant colony, the probability of exploring different paths is given. The probability of the transition path is calculated using the probability combining p1 and p2. And then select the next step to solve the problems that are prone to appear in ant colony algorithm such as:premature convergence and falling into local optimal solutions, or even stagnation.(3) In the early stage of ant colony algorithm, it often takes a long time to explore the path based on the pheromone content left in order to determine the next step of the path direction. Therefore, this thesis tries to redefine pheromone content in the initial period, and then get the distinguish information and thus reduces the initial exploration time. At the same time, in order to avoid the gradual increasing amount of the pheromone content during its initial period that will result in path selection error, and falling into the local optimal solution, this thesis adopts the following information pheromone to update strategy:if the current ant path values are better than all previous optimal paths, then enhance the treatment to the information that the ant itself left on the path; otherwise weaken the treatment.(4) Whether the parameter setting of ant colony algorithm is reasonable or not has a great influence on the performance of the algorithm. This thesis employs eil51 data set and carries out a systematic experimental study on the parameters of the improved ant colony algorithm, and gives the total pheromone content Q, ant number m, pheromone evaporation factor 1-ρ, pheromone heuristic factor a and pheromone expected heuristic factor β parameters and the optimal range.To tackle such problems as discontinuous, smooth ambiguity that are likely to appear in the image edge detection, this paper uses the improved ant colony algorithm to process image edge detection, and gives the specific node pheromone definition, pheromone update strategy (pheromone concentration weakened or enhanced), defines threshold and uses the threshold to do the binary method of pheromone matrix, at the same time with the between-cluster variance to select the edges. Through 15 times of experimenting respectively on the two images in the standard picture library, (the image resolution is 256×256, parameter setting are as flowing:total pheromone content amounted to 1000, number of ants is 256, pheromone evaporation factor of 0.5, pheromone weight 1.0, heuristic factor weights for 3.0.) the experimental results show that this method in image edge detection and thinning degree of smoothness and continuity are better than the Roberts、Sobel、Prewitt、LOG、Canny method.