支持向量机的多目标图像分割算法仿真研究

摘要

研究图像的目标分割和识别优化问题,针对大量自然场景图像中分割出的各种不同日标,从各种背景分割出有意义的目标有困难.为了从图像中准确分割日标,提出了一种支持向量机( SVM)的多目标分割技术.首先将图像分割为较小的区域,并利用区域融合算法将其合并为语义日标.然后通过用户的交互,指定部分关键点和关键区域,并使用支持向量机算法,将图像中的各区域分类为关键区域,融合为最终的日标区域.试验结果显示方法能分割出图像中不同的目标,能够更好地保留图像分割细节信息,使分割结果与人眼视觉的判断标准相近,证明改进的方法能广泛适用于多日标自动识别.%In this paper, we proposed a novel object segmentation algorithm from natural images based on region merging. To do this, we segmented an image into regions and merged them as a semantic object. We created a critical window which contained critical map and critical points from an image. Within the window, a Support Vector Machine was used to select the critical regions, which were then clustered into the objects using the region merging method. The proposed method allows multiple objects to be segmented. Experiments have shown results close to human perception. The proposed method can be used for face detection, flight objects detection, and etc.