基于TWSVM超像素分类的彩色图像分割算法

Image segmentation based on TWSVM superpixel classification

图像分割是图像分析与理解的关键环节之一.提出了一种基于TWSVM超像素分类的彩色图像分割算法.首先,利用熵率超像素生成算法,将原始彩色图像划分成超像素区域;其次,结合直方图与双树复小波变换理论,提取出超像素的颜色特征和纹理特征;然后,采纳最大类间方差阈值法确定出TWSVM训练样本;最后,利用训练好的TWSVM模型对超像素进行分类处理,以获得最终分割结果.实验结果表明,本文算法可以获得较好的彩色图像分割效果.

Image segmentation is the key step of image analysis and image understanding.In this paper,we propose a TWSVM superpixel classification based color image segmentation.Firstly,the original color image is divided into superpixel regons by using the entropy rate superpixel algorithm.Secondly,the color feature and texture feature of superpixel regions are extracted.Thirdly,both color and texture features are used as input of TWSVM model,and the TWSVM classifieris trained by selecting the training samples with OTSU thresholding.Finally,the color image is segmented with the trained TWSVM classifier.Experimental results show that our proposed method has the promising segmentation performance.