Research On Multi-resolution Volume Rendering Of Large Scale Seismic Data

As a technique of 3D data filed visualization, volume rendering has been widely researched and used in many domains because of its ability of showing the inner information of object. With the development of data acquisition and measurement technique, the scale of 3D data field has been larger and larger which is beyond the memory size of CPU or GPU. Parallel rendering can deal with the large scale data set rendering which divide the data set into blocks and parallel render them using several GPU and combine the results into a final image. However, the application condition can hardly satisfy the hardware requirement of parallel rendering. In seismic prospecting field, the scale 3D seismic data set is larger and larger due to the development of prospecting technique. However, the scale of interest target region of seismic data is much smaller compared to the whole data set. Based on that, the multi-resolution method was proposed. It blocks the data set and renders each block with different resolutions. Due to the scale of target region is much smaller than the scale of none target region, it can reduce the whole data amount by lowering the resolution of none target region.In this thesis, we focus on the multi-resolution method of large scale seismic data set and doing research with the consideration of the low signal-noise-ration and fuzzy target region border of seismic data set. Our achievement is below:We introduced an adaptive block rank truncation and resolution selecting method in tensor approximation. Traditional multi-resolution method uses the comentropy to determine the resolution of each block. Different with other filed, the data set in seismic prospecting has the character of low signal-noise-ratio, extreme data variation and much structure information. It will lead to the high comentropy of each block which will cause the loss of structure information in seismic data when guaranteeing the high compression ratio. Recently, 3D tensor approximation has been widely used in multi-resolution volume rendering which using the 3D principal component analysis(PCA) to reserve the structure feature of seismic data in multi-resolution. For the problem of determining the compression ratio and resolution of each block in the tensor approximation based multi-resolution volume rendering, we introduced an adaptive rank truncation and resolution selecting method which can find the best compression ratio and resolution of each block adaptively. The result shows that our method can effectively solve the problem of determining the compression ratio and resolution in tensor approximation;We proposed a low-rank-residual-iteration-based tensor decomposition method. In tensor approximation, there is a paradox that the high rank decomposition has not high enough compression ratio while the low rank decomposition has bad image quality. We proposed a low-rank-residual-iteration-based tensor decomposition method which can lower the low-rank decomposition error by iterating the residual tensor of low-rank decomposition. The result shows that our method can effectively improve the image quality of low-rank approximation when meeting the high requirement of compression ratio;We proposed a GPU-indirect-interpolation-based method to deal with the block artifact in multi-resolution volume rendering. Dividing the data set into blocks is the base approach of multi-resolution volume rendering which will lead to the block artifact in the final image. Previous approach to deal with the block artifact is based on the border data duplication method which requires a lot of extra memory to store the extra data. We proposed the GPU-indirect-interpolation-based method to deal with block artifact which searches the border data of neighboring block in GPU to complete the interpolation while rendering. Result shows that our method can effectively weaken the block artifact without increasing any extra storage memory.In conclusion, we proposed several effective methods to deal with the problem in large scale seismic data multi-resolution volume rendering in this thesis. And our methods have quite high practical value.