Design And Implementation Of Super-resolution Reconstruction System Based On ARM+FPGA

Image super-resolution reconstruction technology can improve the resolution of the observed images via digital signal process under the condition of the existing imaging equipment,which is of great significance in improving the quality of the images.At present,the domestic and international research of image super-resolution reconstruction technology is mainly to complete the design and optimization of algorithms by computer,but the work of transplanting the reconstruction algorithm into the small image processing equipment and achieving real-time super-resolution reconstruction is less involved.For this reason,we design a set of super-resolution reconstruction system,which can realize the off-line process of super-resolution reconstruction in this thesis.The main contents of this thesis include the introduction and selection of super-resolution reconstruction algorithms,the design of the super-resolution reconstruction system,and the hardware implementation and optimization of the super-resolution reconstruction algorithm.The specific work is as follows:(1)We analyze the advantages and disadvantages of the classical super-resolution reconstruction algorithms,and compare the reconstruction results.On this basis,we select the super-resolution reconstruction based on dictionary learning as the reconstruction algorithm of this system.(2)A kind of image reconstruction system based on ARM+FPGA is designed.The system can accomplish the functions of image capture and buffer,the communication of ARM and FPGA,super-resolution reconstruction,and the real-time display of reconstruction results.(3)Considering the features of hardware and software of the system,we design a transplantation scheme of the super-resolution reconstruction algorithm.The principal component analysis(PCA)dimension reduction and orthogonal matching pursuit(OMP)algorithm of the super-resolution reconstruction algorithm are transplanted into the FPGA,which include large amount of computation and can be processed in parallel.The serial operational algorithm is processed by ARM.(4)According to the transplantation of embedded Linux system and FPGA developmentprocess,we program C program and hardware description language(Verilog HDL)to realize the algorithm.And we realize the operation modules such as interpolation and eigenvalue extraction in ARM,matrix multiplication,matrix decomposition and matrix inversion in FPGA.Then we use FPGA modules to complete the FPGA implementation of PCA dimension reduction and OMP algorithm.And on this basis,the system optimization scheme is designed which includes the algorithm optimization,the parallel processing optimization,and the compile optimization.(5)Using the characteristics of FPGA parallel computing,we have completed the structure of multiple multipliers and adders to accelerate the matrix multiplication process in PCA reduction and OMP algorithm,which greatly improves the speed of the image reconstruction.(6)At last,the whole machine is debugged,the validity of the system is verified,and the quality and speed of image reconstruction under different operating conditions are analyzed.