Research And Implementation On 3D Model Retrieval Algorithm Based On HOG And Skeleton

In recent years, with the development of computer hardware and laser scanner technology, 3D models have become the fourth generation of multimedia data types and they are in a research upsurge. 3D models are often applied to computer graphics, computer vision, molecular biology, entertainment and so on. There are many different kinds of large 3D models libraries emerged on the Internet, such as drug 3D models libraries, general 3D models libraries and so on. Rather than modeling a new 3D model, we can find out the wanted or similar ones in the 3D models libraries and do the second modeling to get the best ones we wanted, and it will save much time and human cost. Therefore how to find out the wanted 3D models fast and accurately in a huge models library is a viable research area.3D models have many features and each one has its advantages, disadvantages and applicable scenarios. This paper discusses the retrieval technology with hybrid features to keep the efficiency of retrieval accurately and stably in a large models library. The system provides the retrieval interface with 2D sketch and mixes the topological and local texture features to improve the efficiency of retrieval. The main work of this paper includes:(1) We propose an improved scheme on path similarity skeleton graph matching and apply it to retrieve 3D models. Skeleton graph matching includes path similarity, node similarity, graph similarity. Graph similarity computing uses dynamic programming algorithm, which cannot change the sequence of node while finding the best matching. The algorithm is used to retrieve 3D model, and it can do a fairly good job.(2) We propose a hierarchical fusion algorithm based on HOG and skeleton features. Firstly, this paper matches HOG feature of the sketch with HOG features of all the 3D models in the library. Secondly, it matches skeleton feature of the sketch with skeleton features of top N models in the first matching to improve the order of the result. It plays fully the advantages of local texture and topological feature.(3) We design and implement the prototype system called 3D Model Retrieval System. Based on theoretical research on retrieval algorithm, we implement it in the actual project to prove its effectiveness and design the prototype system to visualize it. What is more, we contribute to the ease of use in the system to improve the user experience.Experiments in this system show that the retrieval result of hierarchical fusion of HOG and skeleton features is better than the single one’s, and it improves the accuracy of retrieval. On the same time, our system compares to the famous 3D model search engine of Princeton University and our system performs better.