中文摘要：

提出了一种在频域空间内用频率谱变化量来表示图像轮廓特征的方法,并在宗教类唐卡图像的头饰分类中得到了成功的应用。标注图像的头饰区域,应用基本全局门限和用户观察直方图选择阈值相结合的方法分割标注区域;将图像中的像素点转换成直角坐标系下的坐标点,并提取典型轮廓点;通过傅里叶变换将典型轮廓点转换到频域空间,应用频率谱值较小的那部分的变化量作为轮廓特征,提取轮廓内部的颜色特征。实验证明,应用所提取的头饰特征可以有效地对唐卡图像进行分类。

英文摘要：

This paper presents a method based on the variation of frequency spectral to express contour feature,which has been successfully applied to classify headdresses of religious portrait Thangka image.Mark the headdress region of the image,segment the marked area by the means of basic global threshold and user observation histogram to choose threshold value.Transfer the pixels in the image into points in the coordinate system,and extract the typical contour points.Transfer the typical contour points into frequency region space through Fourier transform,apply the relatively small frequency spectrum values as the contour features.The experiments show that the effective classification of headdresses can be done by extracting the features of headdresses in this paper.