Three-dimensional Animated Mesh Compression Via Saliency And Spatial Features

In recent years,three-dimensional dynamic mesh sequences(DMS)have been widely applied in various fields such as remote teaching,online games and telemedicine.In order to express the strong sense of reality of three-dimensional dynamic mesh sequences,the detailed information of three-dimensional dynamic mesh sequences is constantly enriched,which leads to the amount of data expands exponentially.Therefore,it is great significance to design an efficient three-dimensional dynamic mesh sequences compression algorithm for the application and development of three-dimensional animation.We proposed a novel processing framework to improve the compression rate of DMS by eliminating redundancy in spatial and temporal,and laying a foundation for realizing progressive tranmission in this paper.In order to realize progressive transmission of three-dimensional dynamic mesh sequences,We proposes a type of temporal frame-clustering algorithm.Firstly,the algorithm divides the classification boundary of DMS by the measurement of local motion similarity.Then,the vertexes belong to the non-classification boundary is clustered with local motion similarity and the key-point is used to represent the clustering region,so as to reduce the cost of calculation.Finally,we transform the temporal clustering into the similarity evaluation of salient value between frames by constructing the salient model of three-dimensional dynamic mesh sequences.This definition is not only more accurate but also greatly meets the user's visual demand in clustering.In view of the above-mentioned temporal clustering results,we calculated the mesh residuals between the key frames and other frames in the same class based on the method which predicting the value of three-dimensional dynamic mesh sequences,and the differential matrix is regarded as the mesh signal.Then,executing spatial subdivision by pondering the continuous of topological in order to enhance redundancy removal efficiency.And we construct three trajectory matrixs in x,y,z direction respectively,and applied the PCA to obtain a set of coefficients which would be regarded as the mesh signal.Finally we apply Graph Fourier Transform(GFT)to produce transform coefficients with the these signals,which are then coded by CSPECK.Experimental results show that our method is superior to other methods in terms of storage requirements and refactoring quality.