Software Design Of JPEG XR Image Encoder Based On Tilera Multi-core Processor

As an image compression standard by JPEG of new generation, JPEG XR has not been researched a lot. The number of applications is much less than that of JPEG and JPEG2000, which has been focused on by this issue. By the combination of the potential research hotspot and the practical application demand, this thesis design and developed JPEG XR encoder based general multi-core processor, TilePro64.Firstly, in order to develop a guideline, this thesis analyzed and summarized the characteristics, methods and related theories of JPEG XR image compression standard. Then, this thesis introduced the background, hardware structure and software development environment of TiIePro64, which indicated that this processor is an ideal carrier of JPEG XR image encoder with certain complexity. This thesis mainly described the scheme of JPEG XR encoder, which intended to optimize two features.One is the running efficiency. The encoder proposed by this thesis has a parallel architecture organized according to the function of each module. It was able to alternate among different pipelines and change the number of cores dynamically. The testing results proved that the scheme improve the running efficiency effectively and the parallel speedup is up to3.The other is compression efficiency. Additional function had been developed to support the lossless coding of Regions-of-Interest with arbitrary shapes in the compressed pictures, with which the Regions-of-Interest could be much clearer under the certain total compression efficiency. The selection algorithm of JPEG XR quantization parameters based on subjective quality assessment proposed by this thesis introduced the features of human visual system in the encoding process. The algorithm mapped the default quantization parameters inputted by users to the actual quantization parameter set. Each region of the image is assigned a set of actual quantization parameters according to its local texture and local brightness, which distributes the bit rate of the entire image reasonably. The compression efficiency of the output image was improved up to10%with the subjective quality remaining unchanged.Finally, after the conclusion of this issue, this thesis simply prospected the further optimization and the future direction.