基于变尺度进化的多目标图像分割算法  

Multi-objective Evolutionary Image Segmentation Algorithm Based on Varying-scale

作　　者：周苑[1] 周岩[1] 

机构地区：[1]河南工程学院计算机学院,河南郑州451191

出　　处：《电视技术》2015年第17期21-24,共4页Video Engineering

摘　　要：针对进化计算在图像处理方面的应用存在计算代价高和评价函数单一的缺点，提出了一种变尺度进化的多目标图像分割算法。该算法在进化的不同阶段将不同数量的相邻像素作为同一像素处理，从而在不同尺度下完成图像的分割，降低了算法的计算时间。同时，设计了监测关键解变化量的方法采控制不同尺度的过渡时机。在评价函数上则采用包含紧致度和连接度的多目标方式，以更为全面地评估分割的质量。其中，连接度检查邻接像素所属类别的连续性，有效抑制了噪声和小样本类的干扰。对Sailboat和Terra图像的分割结果显示，所提出的算法能有效地将不同对象分割开来，同时在计算时间上仅为几种同类对比算法的48．08％～55．53％。Ewolutionary computing has some shortages such as high computational cost and single measurement when it is used im the field of image processing. Hence,a novel multi-objective ew）lutionary image segmentation algorithm based on varying scale is proposed. ＇Fire algorithm regards differcnt regions of image as one pixel in different stages of evolution so that the image is segmen- ted in different scales and the total consumed time is reduced. The transition moments among different scales are controlled by mo- nitoring the variation of important solutions. On the other hand of measurement, a muhi-objeetive version of compactness and con- neetivity is used to well estimate the image＇ s quality,wherein the eonnecfivitv is modified by ehecking the continuity of clusler la- bel among neighboring pixels. This modificalion can effectively suppress the influence of noise and small clusters. Experimental re- sults on standard images of ＂sailboat＂ and ＂lerra＂ show that the proposed algorithm can well distinguish different patterns in the images and the time consumed is remarkahle reduced to 48.08% ~ 55.53% of those in other algorithms.

关 键 词：进化计算 多目标 图像分割 变尺度 连接度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]