【摘要】 为实现对数字虚拟人彩色连续切片的自动分割,提出了一种基于支持向量机的自动分割方法.该方法首先利用虚拟人连续切片图像间在颜色、纹理等特征的相似性,使用支持向量机分类方法对相邻图像进行分割.在分割结果达不到预定允许最小相似度条件或一次训练结果用于连续分割切片次数达到给定的最大允许次数时,则再利用相邻切片在空间上的相似性,自动对当前切片进行重新采样,然后继续训练、分割.使用虚拟中国人女性一号(VCH-F1)数据集中切片图像进行实验,对比研究了允许的最小相似度测度及允许最大连续分割次数对分割质量和分割时间的影响,并给出了受噪声污染图像的自动分割结果.该方法能有效实现对虚拟人连续切片图像的自动分割,能在保证一定分割质量的同时,兼顾分割速度,并能较好地克服噪声影响.更多还原

【Abstract】 For segmenting the slice images of the virtual human dataset automatically,an auto-segmentation method based on support vector machines(SVM) was proposed.A similar measure was defined to detect the effect of the auto-segmentation.The neighboring slice images were segmented by the classification of SVM by using the similar characters in color or texture between neighboring images.The image would be re-sampled by using the similar spatial characters,if the segmentation result can′t satisfy the specified minimum similar measure,or if the segmentation times reached the specified value that presented the allowed continuous maximum segmentation times after one train.The method would continue to segment the current image and the next several images.The experiments were carried out by using the slice images of virtual chinese human No.1 female(VCH-F1) Dataset.A compared result on segmentation quality and segmentation speed was shown with the minimum similar measure and the allowed maximum segmentation times of one train,and a compared segmentation result of noisy images was given.The method can auto-segment the color slice images of Virtual Human Dataset effectively,with a satisfied segmentation quality and a suitable time cost and can overcome the effect of the noisy images.更多还原