基于GPU的小波变换

Wavelet Analysis Based on GPU

针对图形处理器研究问题,其中图形海量数据集的分析与处理,多用小波变换方法。但计算量大,难以适应实时性要求。近年来图形处理器的性能大幅提高,其深度流水线和并行运算机制提高,为解决实时计算问题提供了良好的平台。在研究小波变换矩阵形式及GPU编程模型的基础上,提出了一种关于GPU的小波变换方法,利用数组与纹理之间的对应关系实现小波变换,将离散的数据点映射到纹理,将小波变换的计算影射为高维矩阵与向量间的乘积形式,并通过渲染到纹理的形式取得中间结果。方法充分发挥了GPU流水线的并行性优势,实验表明方法可有效减少计算时间,从而达到实时绘制的要求。

Wavelet Analysis has been brought into wide use in large scale data processing,but it can not fit the demand of real-time simulation because the quantity of computing is vast.In recent years,the capability of GPU is improved greatly,the parallel pipeline processor and programmability of GPU provide a solution to this question.We present a method for wavelet analysis based on GPU and matrix arithmetic of wavelet,which uses the matrix and texture to complete the computation,transforms the data to texture,transforms the wavelet analysis to product of the matrix and vector,and uses the rending to get the results of computation.The method takes advantage of GPU pipeline,and the results of experiment indicate that the method reduces the time of calculation effectively,so that it achieves the demand of real-time render.