Portable Network Graphics Decoding With GPU Accelerating

PNG (Portable Network Graphics) is a scalable image format that employs lossless datacompression. PNG has alpha channel transparency and supports palette-based images,grayscale images and truecolor images. PNGs are allowed for a maximum color depth of16bits. Compared with BMP, JPEG, GIF and other image formats, PNG is characterized by thelossless data compression and alpha channels, and therefore is widely used in website design,video&image composition, data compression, and software user interface design, and etc. AsPNG file is created by serial encoding, it is decoded in a lower speed. However decodingspeed is critically important when PNG sequence is used in real-time video compositing orreal-time display of images in batch. Thus, decoding speed has becomed a bottleneck in PNGfile application. An immediate accelerating solution is in demand.NVIDIA unveiled CUDA (Compute Unified Device Architecture) in November2006, afundamentally new architecture for computing using GPU (Graphics Processing Units).CUDA is developed by imitating C language model, which shed some light on the approachof solving general purpose problems using GPU. CUDA is now widely used in astronomicalcomputing, biological computing, image processing, audio&video codecs and many otherfields. It runs up to a hundred times faster in application. This dissertation discusses PNGdecoding using NVIDIA GPU for the first time, and mainly studies parallel algorithms foraccelerating PNG decoding. By analyzing the elapse time of each process in decoding withGPU, this dissertation first proposes a method to accelerate parallel computing in thede-filtering process which costs more than50%of elapse time during decoding. Thedissertation also comes up solutions to optimize the algorithm. Secondly, this dissertationdevelops a batch processing parallel algorithm for de-filtering process, which speeds up thedecoding further. At the end, the dissertation draws up a PNG decoding process diagram, inwhich process GPU is used in parallel with CPU. A PNG sequence player is developed on thisbasis, and it proves that parallel computing and decoding acceleration discussed in thedissertation is technically practical in application.