Research On Image Super-resolution Reconstruction Technology Based On Embedded System

As digital imaging technology is widely used in satellite remote sensing, bio-medical, public security and other fields, the requirements of digital image quality are getting higher and higher, and the resolution of existing imaging systems can't meet people's demand. Only by appropriate digital signal processing method, can the super-resolution(SR) technology improve the image quality without changing existing image systems. This technology has the advantage of cost, and has been widely concerned in recent years. However, most of the SR techniques cannot accomplish image reconstruction without the help of a computer. In other words, they cannot achieve the real-time processing, which greatly limits the application of this technology. Therefore, it is of great practical significance to combine the SR technology with embedded systems and develop miniaturization and portable image processing products.The research object of this paper is to design a SR reconstruction system which can achieve image quality improvement of offline processing. The main research content includes: the hardware platform design, the selection and the improvement of SR algorithm, the hardware implementation of the algorithm and the optimization of the system. The specific work is as follows:(1)Design a kind of embedded image processing system based on ARM+FPGA, which with ARM as the control center and FPGA as the data processing center, has a good system management capabilities and excellent data parallel processing capabilities to meet the hardware requirements of image processing.(2)Analysis of the advantages and disadvantages of existing SR algorithm. According to the characteristics of embedded systems, select a learning-based SR algorithm which is suitable for hardware transplant as the reconstruction algorithm of the system.(3)Study the transplantation method of Linux to embedded, and introduce how to build a cross-compiler environment, and the transplant process of u-boot and Linux kernel is given.(4)The principle of SR algorithm based on sparse representation is studied, and some improvements are made to this algorithm, which include using the joint dictionary training method, establishing multi-dictionary based on classification, and optimizing the image feature extraction operator. Experimental results show that these improvements can improve image quality and reduce the time of the algorithm running.(5)According to the characteristics of the system software and hardware, the transplant program and process is given. The OMP algorithm, which is able to parallel processing and has a large amount of computation, is achieved with FPGA and the rest of SR algorithm is done with ARM. On this basis, some system optimizations are given, including algorithm optimization, compiler optimization and hardware system optimization.(6)The whole system construction and debugging is completed, and we achieve the image SR reconstruction with this system, which can verify the effectiveness of this system.