Feature Extraction Technologies Research And Implementation Of 3D Models Based On Wavelet Transform

With the emergence of a variety of modeling techniques based on three-dimensionalmodel,three-dimensional model shows geometric growth. If the existing three-dimensionalmodel can be better and more reasonably reused, a great deal of resources and manpower willbe inevitably saved. Thus, the problem "how to build three-dimensional model?"is transferredinto "How to check and reuse three-dimensional model?". This paper conducts a study on theextractive technique of three-dimensional model based on its shape for solving the problem ofhow to query the three-dimensional model.The key problem of retrieval of Three-dimensional model based on object shape is quickand accurate extracting features of three-dimensional model. Researchers home and abroadconstantly propose and improve the feature extraction algorithm. After reading a lot ofliterature, this paper makes improvment based on G. Burel et al who proposed a ray castingalgorithm and D.V Vranic who proposed to improve the light projection. G. Burel ray castingalgorithm is simple, but for complex model, only retaining the far intersection point of modelsurface -ray ,which will lose a lot of geometric information and the accuracy of featureextractionwill be reduced. Vranic proposed multi-sphere light projection algorithm based onG. Burel, but the stability of the algorithm is not strong. So this paper proposes a featureextraction algorithm, which is based on spherical wavelet transform feature. We obtain theintersection of ray-model surface by using Chebyshev sampling points , using the discretewavelet transform to obtain multi-scale pyramid decomposition of sequence of Chebyshevsampling points. Finally, the invariance of spherical harmonic transform is used to obtainrotation invariant feature vector. To verify the feasibility of the method, a visual search featureextraction experiment system was developed. This system has realised the display ofthree-dimensional model of triangular patches and voxels , normalizing the pretreatmentmodel, and can intuitionally see the visual effect of various retrieval effectiveness , using thePrecision-Recall (ie: recall - precision) curve to evaluate the proposed algorithm. Throughvarious methods of experimentation and evaluation, the overall model geometric informationcan be well described through formed feature values by using spherical wavelet transform,meanwhile, the characteristic of the details of the model is well described according to amulti-level description of the characteristic of the shape of the model, and the noise remainsstable.to a certain degree.