摘要
为了加快用于图像分割的支持向量机算法的训练速度,本文提出主动选择样本简化训练集的新方法。该方法根据像素在颜色空间的统计特性构建可分的训练集,并采用均匀抽样策略大大缩减训练集规模而不降低分类正确率,使得支持向量机可以实时训练,并为参数调整带来便利。由此发展了一种非监督算法与支持向量机相结合的自动图像分割方法。通过支持向量机在线训练,新方法可以获得较高的分割精度,有较好的鲁棒性,现已应用于彩色血细胞图像分割。
In order to speed up the training of SVM for image segmentation, two novel approaches for active sample selection are proposed. According to the statistical property of pixels in color space , one is making the training set separable , and the other is uniform sampling from the original data. The size of training set can be reduced significantly and the accuracy of classification is not decreased by those approaches. It makes training of SVM completed in real-time and brings convenience to adjust parameters. So a mixed method that combines unsupervised method with SVM is developed to segment images automatically. By training SVM on-line, the new method can achieve high accuracy and is robust to varied image acquisition. Experimental results demonstrat that our algorithm is suitable for segment blood cell images.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2005年第4期392-398,共7页
Pattern Recognition and Artificial Intelligence
关键词
彩色图像分割
支持向量机
训练
参数调整
Color Image Segmentation, Support Vector Machine, Training, Parameter Adjust