Research On Frescos-based Object Recognition Technique

Dunhuang Cultural Research is one of the most popular fields in the preservation of cultural relics at present. It's bearing mostly to study intelligently on Dunhuang frescos using computer technique. This paper is dedicated to the research of high resolution Fresco based object recognition. It studies how to recognize kinds of objects from a high-resolution fresco. It aims to release the researchers from recognizing through their eyes. It also presents a method for establishing semantic database on frescos and researching relatively.This paper summarizes the basic theories and methods on object recognition, and introduces the local feature descriptor which is used widely in object recognition. An invariant local feature descriptor called SIFT is applied to recognizing objects from frescos in this paper. High resolution fresco oriented object recognition is separated into two aspects. First, in order to resolve the problem that high resolution image recognition require highly system, the paper studies how to access the image quickly, presents a multi-scale method of large-scale image storage, and finally completes a view tool to large-scale image. Second, this paper studies on SIFT based object recognition of the reduced images from above multi-scale image architecture.This paper studies thoroughly the procedure of frescos oriented object recognition which is based on SIFT, and a prototype system is developed according to the procedure. To resolve the SIFT' insufficiencies in frescos-based recognition, this paper also provides an improved feature descriptor which adds the shape context to SIFT descriptor.