Two-Stage Thangka Image Missing Key Block Retrieval By A Combination Of Semantic Feature And Visual Feature

Up to the present, the research of the image retrieval has undergone three stages. Thangka image, the treasure of the Tibetan people, scholars have achieved much production on the research of Thangka image retrieval domain, but the research on the Thangka without key blocks just lies in the beginning phase. So, it is an important and challenging job to find a way to retrieve the user-wanted images from the image databases rapidly and effectively.On the basis of an overview of retrieval methods based on semantic and content, comminuting with characteristic of Thangka image, taking images missing key blocks as main research subjects, this paper put forward two-stage Thangk image retrieval by a combination of semantic and visual feature, and we implement a prototype of this approach. The thesis aims to finish the work as the following.First of all, classify Thangka image to several categories, which dwindle the range of second retrieval, improve the retrieval precision. It is because the Thangka missing its sole key block, we cannot extract features of these missing blocks, it is necessary to use other key block, as classification key block, to reflect the characteristic that not existing. Then extract the features of classification key block, achieve its annotation and the first-stage image retrieval.Secondly, the segmentation and visual feature extraction of region surrounding with missing key blocks. On the basis of image connected components labeling algorithm, this paper proposed a Thangka image segmentation. Segment the region surrounding with missing key blocks, then extract its features and second stage retrieval.Finally, on the basis of two methods are combined and applied, achieve a Thangka image missing sole key blocks retrieval system, then calculate according precision and recall, and analysis the experiment results.This paper realizes a two-stage Thangka image missing key blocks retrieval system with a combination of semantic and visual feature. The simulation results show that this retrieval system needs the users’ requests, and increases the users’ satisfaction.