Frequency Domain Real-time Image Processing Based On FPGA

Real-time image processing system is widely used in military occasions.Real-time image processing algorithms are generally of great data size, highcalculating complexity and high real-time feature. Therefore, PLDs(programmablelogic devices), especially FPGAs(Field Programmable Gate Arrays) are suitable inreal-time image processing systems due to their high performance, high flexibility andhigh speed.According to the functioning scope, image processing algorithms are divided intotwo categories, time domain (space domain) and frequency domain. Although themost commonly used algorithms are all time domain algorithms, in some specialapplications such as image compression, image enhancement, image restoration,image merge and frequency domain filtering, frequency domain algorithms haveevident advantage over time domain algorithms. FFT is one of the most commonoperations in digital signal processing, and many basic operations can be transformedby means of FFT and calculate more easily. As to image applications, FFT operationis extended to two-dimension, and some fundamental calculations are FFT related, forexample, correlation, filtering, spectrum estimation and convolution can be easilytransformed into FFT operation to calculate, thus research on two-dimensionFFT(2D-FFT) is of some certain significance. This paper focus mainly on thereal-time implementation and application of2D-FFT。This paper analyzes the theoretical method to reduce the dimension, and thenbrings out the pipelined2D-FFT according to the hierarchical top-down designmethod. All modules are implemented by means of the hardware descriptionlanguage(VHDL) on the image processing hardware platform which is designed with the Xilinx Virtex-V FPGA. A new way to arrange the butterfly unit into the pipeline,thus solving the problem that the subtraction item is not continuous, is proposed, anda series of optimization suggestions which can be used to save the resources andspeed up the calculations are proposed, too. Based on the2D-FFT, we bring out thecommonly used image interpolation operation in both time(space) domain andfrequency domain to make some comparison, and then implement the imageself-filtering in frequency domain, which is very hard to realize in the time(space)domain. The result of synthesis shows that, the maximum frequency of the algorithmdescribed in this paper reaches335.593MHz.In order to verify the accuracy of the2D-FFT described in this paper, functionalsimulation is designed, and the results are compared with that of the Matlab, and thenwe implement the algorithm on the hardware platform, of which the result issatisfying, representing the advantage of the real-time image frequency domainalgorithm described in this paper. The work described in this paper reaches somecertain accuracy, and can fulfill the practical projects. It is of some reference value forthe application of the real-time frequency image processing algorithms and advancedresearch, and it is of some applicational value, too.