Study On The Algorithms Of High-speed Image Processing Based On FPGA And System Implementation

Image processing is still facing many challenges featuring networking of image processing, solving of complex problems and high requirements on processing speed. High-speed image processing systems based on field programmable gate arrays (FPGA) are the primary means to rise above these challenges. This paper researched the key technologies of FPGA-based high-speed image processing systems. A FPGA-based high-speed high-definition medical image processing platform and a FPGA-based presswork detecting platform were implemented.Firstly, the design principle is presented that synchronous pipelines are suited for FPGA-based high-speed image processing; the timing constraint formulas for algorithm modules of pipeline-based image processing are given; a method is proposed to optimize FPGA-based pipelines in accordance with the structural characteristics of FPGAs. The optimization method proposed can enhance the performance of synchronous pipelines effectively.Secondly, this paper also presents a time division multiplexing (TDM) method based on buffer-shift algorithms to realize the TDM of shared resources in FPGAs. In this method, the relationship between the displacement requirement of a multiplexer and buffer storage resources is analyzed to facilitate the allocation of these resources. A model of the demand function of a multiplexer is built, and the impact of displacement on demand functions is analyzed. A shared resource allocation method based on time unit is presented to facilitate the realization of TDM scheduling. A shift method based on the curves for the statistical distribution of multiplexers'demands is given to implement the displacements required by multiplexers in a simple and convenient manner. A scheduling mechanism based on state machines is offered and the differences of two kinds of multiplexers in rotation intervals are analyzed to avoid erroneous calculation in a time unit. A circuit was designed to implement the buffer-shift algorithm, and with discussions of fan-in and fan-out, related solutions are given.Thirdly, this paper also proposes a parallel CORDIC algorithm based on maximum approximation angles to solve the problem of implementing in FPGAs the transcendental functions of image processing algorithms. The proposed CORDIC algorithm is advantageous in terms of iteration speed. Then, this paper also presents a two-step method to realize angle conversion via the Bayesian binary regression method and the MAR algorithm. This two-step method solved the computing and storage of the scale factors in the proposed parallel CORDIC algorithm, saved storage units, and speeded up iteration speed. The methods for the hardware implementation of the proposed CORDIC algorithm, especially the adders in the structures, are given.In addition, it's proposed that the speed of printed matter can be measured indirectly through image registration. This method solved the problem of measuring the speed of printed matter in a weak detection environment. It's also proposed that PID controllers based on neural network algorithms can be used to drive linear array cameras, solving the problem of the driving of linear array cameras in a weak detection environment. The approaches for implementing neural networks that are based on time sequencing constrains and FPGAs are given.Finally, based on the above theoretical results, a proprietary FPGA-based high-speed high-definition medical imaging platform and a FPGA-based presswork detecting platform have been developed according to the requirements from certain companies. The platforms have been successfully applied to production practice, bringing out good economic and social benefits.