A Study On Automatic Annotation Of Thangka Image Based On Graph Learning

This paper summarized relative techniques of image annotation and proposed a new annotation algorithm which suited for Thangka image. The automatic image annotation of Thangka headdress area was realized by using multi-feature combination method and applied the framework of graph learning annotation algorithm. Specifically, the content of this paper included three aspects. Firstly, a Thangka image annotation sets was built through a lot of research and statistic work among mass of image. This annotation sets provided sample and standard for the digital research of Thangka. Secondly, graph learning annotation algorithm was applied to the area of Thangka annotation area which based on the thought of ROI(Region of Interest). This algorithm was an ideal framework for Thangka image annotation. Thirdly, this paper realized whole annotation process of Thangka headdress area. Customers could obtain the final annotation result without any interactive manipulation but input a Thangka image. Thereinto, saliency map detection algorithm which based on visual focus mechanism was used in the process of image segmentation. Meanwhile, multi-feature combination method was used to extract features and automatic annotation of graph learning framework was applied in the annotation stage.As a exploratory research of Thangka image annotation, this paper fully took the comprehensive and robustness of algorithm into account. The automatic image annotation based on saliency area detection had nice translational and rotational invariance, so that it suited for the head area of buddha which has excursion and swing. And in result of recognition, system had false acceptance function and distinguished it from the normal annotation result. Graph learning annotation algorithm built research framework for Thangka image based on multi-ROI and offered convenience for the research in future. Meanwhile, the final recognition result was satisfy us according to the experiment.