基于头饰特征的肖像类唐卡图像自动标注

人机交互选择肖像类唐卡图像的头饰区域,采用最大类间方差法取得阈值对图像进行分割,得到初始的唐卡分割图.计算头饰区域的欧拉数、头饰轮廓内部的颜色分布,综合这两个维度的特征刻画头饰的整体特征.头饰特征到肖像类唐卡图像名称具有对应规则.结合这一规则,实现肖像类唐卡图像的名称标注.实验结果表明,整合多特征,利用底层特征到名称的映射,可以有效地对肖像类唐卡图像进行标注.

Automatic Annotation for Religious Thangka Images Based on Headress Feature

Artificial selection the headdress region of the Thangka images is choosed using the method based on the maximum between- cluster a best threshold value, then is divided the portrait Thangka image. Euler number and the color distribution value within the headdress profile was calculated. The whole characteristics of the portrait Thangka were described by using the two dimensional properities. There was a certain relations between portrait Thangka＇ s name and headdresses characteristics of religious portrait Thangka. We can annotate the portrait Thangka＇ s name with the relations in order to acquire effective experimental results.