The cluster computer based parallel rendering is an effective method to improve graphics system performance and resolution, the image depth fusion is the key step of sort-last parallel rendering system. This paper puts forward a set of strategies of image depth fusion in sort-last parallel rendering system. Regarding this feature of image depth data, this strategy puts forward template based depth information LZW non-destructive compression method and image depth information interpolation method; regarding the feature of image color data, it puts forward Jpeg based image color information destructive compression method; regarding the image depth synthesis step, it gives a transparence based fusion method. The experimental result shows that this strategy is better than other compression based image synthesis strategies if it is used in sort-last parallel rendering system.
In the research field of virtual reality technology, the traditional method of high performance and high resolution scene rendering is to use large scale professional graphics processors, e.g., SGI graphics processor, while with the introduction of parallel rendering concept, the rendering system has made rapid development. The research on PC cluster based parallel rendering system has become the key method constructing graphics system of high performance and high precision requirements. In 1994, Molnar and others divided the parallel rendering system into three types according to the relationship between the stages of parallel task division and typical graphics rendering flow: Sort-first, sort-middle and sort-last. Tasks of sort-first system are divided according to screen space, however, the uneven distribution of primitives on screen and difference of complex procedures of primitives easily leads to the imbalance of system load; sort-middle is usually applicable to hardware realization; while sort-last parallel rendering system does not divide graphics space, each rendering processor is independent mutually and redistribute pixels till the end of rasterization stage.
Sort-last parallel rendering system causes little unbalanced load due to the uneven distribution of scene data, while increases the processing steps of image fusion. The transmission and fusion of image is the main overhead and bottleneck of sort-last parallel rendering system, so how to shorten the transmission time of image data and optimize the fusion result of image has become an important issue of sort-last parallel rendering system.