Research On Color Video Super-resolution Reconstruction Algorithm And Hardware Implementation

Super-resolution reconstruction is a kind of image processing technique, which can gainhigh resolution image from several low resolution frames with complementary information. Itbreaks through imaging system’s limitation of low resolution, and enhances image resolutionby software technique.By utilizing super-resolution reconstruction, increased resolution canbe achieved without updating the existing low-resolution imaging device, and limitations ofimproving image resolution created by hardware technique such as high cost, high difficultyand limited improvement can be avoided. This paper focuses on super-resolution algorithmand covers such content as super-resolution algorithm of color video based on multi-frames,pre-processing technique of super-resolution reconstruction and the implementation ofalgorithm in video signal system.In the respect of super-resolution algorithm of color video, the covariance matrixestimation of3DISKR algorithm is optimized by several loop iterations, as a result, gain moreaccurate adaptive kernel estimation, and we also improve the times of iterations in motionestimation, regularization parameter and structure sensitivity parameters. The optimizedalgorithm was simulated by using Matlab to process the video sequences Calendar andFoliage, and obtain peak signal to noise ratio, structural similarity and visual effect.According to the analysis of reconstructed result, optimized3DISKR is able to get betterreconstructed effect. In addition, for color image super-resolution reconstruction, this paperproposes a RGB color image reconstruction algorithm based on luminance compensation,which adopts the luminance information of super-resolution construction z_Y(k) as aconstraint, and enables the luminance component y_Ycompensatory(k) after compensationapproximate z_Y(k), thus gains compensatory amount of information, and a colorized highresolution sequence will be composed.In the respect of hardware implementation, first of all we realize image enhancementalgorithm on the platform of Cyclone V5CGXFC7D6F31C7N FPGA. According to the boardvalidation, the effect of processing can get a satisfied result. And then we realize videosuper-resolution algorithm on the C674x-DSP, hardware debugging is carried out by theemulator on the environment of CCS. Finally, we get nice effect by comparing the results, and improve the processing speed of the algorithm significantly by using the DSP hardwareacceleration.