基于杜鹃搜索和二维Fisher准则的图像分割方法

Image segmentation approach based on cuckoo search algorithm and 2-D Fisher criterion

阈值法是图像分割最为常用的方法之一,然而基于一维直方图的阈值方法分割结果容易受噪声的影响.基于二维直方图的二维Fisher准则能够克服一维阈值法缺陷,具有较好的分割性能.但是二维Fisher准则阈值法在求取最优阈值时需要大量的计算,运算速度非常慢.常用的二维Fisher准则阈值优化计算方法如粒子群算法和遗传算法容易陷入局部最优.杜鹃搜索算法是新近提出的一种元启发优化算法,一些经典的函数优化问题测试结果表明杜鹃搜索算法全局寻优能力优于粒子群算法和遗传算法.在介绍杜鹃搜索算法的基础上,提出一种基于杜鹃搜索算法改进的二维Fisher准则阈值分割方法.实验结果证明,提出的方法降低了基本二维Fisher准则阈值法最优阈值的寻找时间,提高了图像分割的实时性,是一种性能良好的图像分割方法.

Thresholding method is one of the most common methods for image segmentation, however thresholding methods based on 1 - D histogram are easily ruined by the noise. Thresholding based on 2 - D histogram and Fisher criterion function can overcome the shortcoming of 1 - D threshold method, which has the better segmentation performance. But due to huge computation is required for 2 - D Fisher criterion function its speed is very slow. Commonly used optimization methods to speed up thresholding based on 2 - D Fisher criterion function like particle swarm optimization and genetic algorithm are easily to fall into the local optimum. Cuckoo search is a newly proposed meta- heuristic optimization algorithm; testing results on some benchmarks indicate that cuckoo search has better global convergence ability than particle swarm optimization and genetic algorithm. In this paper, by employing cuckoo search algorithm, a segmentation approach was proposed based on 2 -D Fisher criterion function. The experimental results show that the proposed method decreases the seeking time of optimal threshold with the basic 2 - D Fisher criterion based thresholding method, which is a well performing method and is more suitable for real -time image segmentation.