| Text location and extraction in natural scene images is a important branch in computer vision field. With a clear application direction and research value, it has become a hot research topic in the world. In this paper we have study on image segmentation based on bacterial foraging optimization algorithm. To achieve accurate and efficient image segmentation, we comprehensive utilize classic and advanced technology to propose a method a new self-adaptive and robust segmentation method with bacterial foraging optimization algorithm as optimization tool.Based on existing research and algorithm, we have proposed several methods to solve some problems on text location and extraction in natural scene image, including the following aspects:(1) Pulse Coupled Neural Network is widely used in image processing field, but it is difficult to define the relative parameters properly in studying the theories and applications of PCNN. So far the determination of parameters of its model needs lots of experiments. To deal with the above problem, a document segmentation method based on the improved PCNN is proposed. It uses the maximum between-cluster variance function as the fitness function of bacterial foraging optimization algorithm, adopts bacterial foraging optimization algorithm to search the optimal parameters and then eliminates the trouble of manually setting the experiment parameters.(2)Based on the analysis on the principle of standard bacterial foraging optimization algorithm, we have improved the bacterial foraging optimization algorithm according to defect in chemotaxis, reproduction, elimination and dispersal. We have designed a adaptive step chemotactic factor and a weight coefficient matrix which is multiplied by the value of the fitness function during the reproduction. The improved algorithm increase the convergence rate and the global search ability.Experimental results show that both precision and recall rates have increased in the improved text location and extraction algorithm. |
PDF Full Text Request