| As the increasing demands for 3D model in industry,animations and games,there are many more ways to acquire 3D model.Both the internet and many other industries have booming increased numbers of 3D model.3D model retrieval system is built for this consequence.The system helps designers in browsing current model,and make other designers life easier.The core of 3D model retrieval system is extracting model features and to quantify the similarities.Current 3D model can divide into two categories: rigid 3D model and non-rigid 3D model.Rigid 3D model has been well researched for many years.In contrast,non-rigid 3D model is a very newly launched topic.With the difference of the criteria to classify rigid 3D model and non-rigid 3D model,the current feature extraction method of rigid model can not be used in non-rigid model.Comparing to rigid 3D model,non-rigid 3D model always have posture changed,and it brings difficulties in correspondence of non-rigid 3D model in database.There are a few non-rigid 3D model extracting methods,but they all have some sort of disadvantages.This paper has discussed how to simply model in advance,which can mostly reduce the numbers of vertices and faces.It will also use the partial model correspondence to present the whole combined with model segmentation based on the spectral descriptors.In the process of segmentation,the result of segmenting is lack of stability and accuracy.So an interest point detector is proposed in this paper.It can improve the segmentation stability with the interest point as the initial clustering center.The final segmentation is assignment by probability distribution of vertices in multiple segmenting.This can improve the accuracy of model segmentation.The hausdorff distance is put forward to realize the model descriptor correspondence.In the end,this paper implements the entire process of model correspondence and completes a non-rigid 3D model retrieval system.The contrast experiments and the display of retrieval system show that the methods adopted by this paper can well fulfil the non-rigid model retrieval request. |
