Three-dimensional Model Of Adaptive Classification Method Based On Feature Fusion And Implementation

In order to find the model in massive quickly, based on independent model library, we have studied and proposed a new Theory that based on the integrated features of the clustering analysis.Designed and implemented a adaptive classification capabilities of the retrieval system for the three-dimensional model,which provided a platform for expansion of the model database and the data optimal management.This research work is supported by 863 Foundation of China; the main contributions of the thesis are as follows:1)Implementation of the model skeleton extraction algorithm, shape histogram algorithm, the model projection circumcision histogram algorithm to describe a three-dimensional model of the classic features. We compared and analyzed these types of algorithm performance,conditions which for further study.2) Based on presented an algorithm on fast transformed a projection point into image, combined with sophisticated two-dimensional image retrieval and relevance feedback methods used, and further presented a dynamic integration feature based on zernike moment features and projection circumcision features extracted from three projection images. Experimental results show improved integration feature of a good effective description of the general model.3) Implementation of the OPTICS, K-Center,CURE algorithm, according to the features data of large and high dimension, presented the entropy measure feature selection algorithm applied to the OPTICS clustering analysis,which improve cluster efficiency also improves the clustering accuracy.4) Based on presented model Library classification rules, established of experimental library. Through study the Princeton library classification, and according to knowledge of cognitive psychology, initially designed a model of classification rules, and based on this to pick out a certain model to build up experimental library.5) Developed a retrieval system with the function of classification of three-dimensional model which can classify a single model, volume model to storage also can be classified to storage, while providing a manual adjustment to change classification error.