基于改进人工蜂群算法的多阈值图像分割

Multi-threshold Image Segmentation Based on Improved Artificial Bee Colony Algorithm

图像分割在模式识别以及机器视觉方面起着至关重要的作用,是图像分析和识别的首要任务。但若分割后图像质量损失严重,就会导致图像后续分析的误差增加。为了能够弥补这一缺陷,在分析了DIRECT算法和人工蜂群算法的特性后,利用DIRECT算法全局收敛并可以快速定位到最优值所在区域的特点来改善人工蜂群算法的过早收敛以及局部搜索能力差的缺点,提出了一种基于改进的人工蜂群算法的多阈值图像分割技术。首先,DIRECT算法为人工蜂群算法提供一种良好的初始种群,种群在演化数代后得到的当前最优解加入到DIRECT算法分割区域中,再进行初始种群的筛选,重复这个过程进而获得最佳阈值并对图像进行分割。为了验证该算法的优劣性,使用峰值信噪比、结构相似性以及特征相似性作为图像质量评价指标并与前人得到的结果进行比较。实验数据表明,提出的阈值分割方法优于前人的阈值分割方法。

Image segmentation plays an important role in pattern recognition and computer vision,which is the primary task for image analysis and recognition.However,it will increase errors if the quality of the segmented image is seriously lost.To solve this drawback,the multi-threshold image segmentation based on improving artificial bee colony algorithm is proposed after analyzing the characteristic of DIRECT and artificial bee colony algorithm.Artificial bee colony algorithm has weakness of premature convergence and poor local search capability,while DIRECT algorithm can improve its deficiencies.The DIRECT algorithm can find a good initial population for artificial bee colony algorithm,and then the current optimum solution can be obtained and joined into the DIRECT’s partitions after several generations of evolution of the population.Keep repeating this process until the stop condition is satisfied.To verify the validity of the proposed algorithm,we adopt the peak signal-to-noise ratio,structural similarity and feature similarity as image quality evaluation indexes and compare with the results obtained by predecessors.The numerical results show that the proposed algorithm proposed is better than the former algorithms.