Research On Bayesian Optimization Algorithm And Its Application In Image Segmentation

Estimation of Distribution Algorithm (EDAs) that incorporate probability model into optimization algorithms have become a new kind of optimization algorithm. It uses statistical learning to build a probabilistic model, and utilizes the model of the sample to achieve the evolution of populations. Bayesian optimization algorithm (BOA) is a representative one among EDAs. It has the advantages of accurate positioning, and can avoid linkage problem efficiently. However, the introduction of statistical learning will bring about a new time and space computational complexity. Bayesian optimization algorithm has been limited in engineering practice because it requires not only a priori knowledge, but also expensive computational complexity in building a probabilistic model.The core of BOA is Bayesian network, and its computational complexity is also concentrated in the construction of Bayesian network. In order to decrease computational complexity, an improved Bayesian optimization algorithm based on immune algorithm is presented. This algorithm intends to decrease computational complexity of BOA by reducing the number of construction of Bayesian network. The immune algorithm by simulating the body's immune mechanism can take advantage of a priori knowledge of the problem and local features to guide the optimization process and improve the convergence speed. Therefore, immune algorithm is introduced into BOA. By utilizing the immune algorithm's guided mutation, the solutions generated by Bayesian network are improved, and the construction times of Bayesian network is reduced. Experimental results show that by compared with the BOA, the proposed algorithm can reduce the computational complexity efficiently and shorten computational time, and has stronger ability of optimization.In order to solve the problem of local optimum in image segmentation, improved Bayesian optimization algorithm is introduced into image segmentation. It uses the optimization capability of BOA to seek optimal image segmentation threshold. The proposed algorithm uses the Bayesian network to encode the pixel, and makes use of Bayesian network sampling to generate a new pixel value. It uses the maximum between-class variance method to determine the fitness function, and searches for the best solution as the optimal segmentation threshold. Simulation results show that by compared with the GA, the improved BOA algorithm have a better image segmentation results. In the present, the papers of application of Bayesian optimization algorithm in image segmentation are published at home and abroad. The method proposed in this paper not only to extend this algorithm's application field, but also for image segmentation to find new solutions.