超分辨率卷积神经网络算法在高速摄影图像优化中的应用

Application of Super-Resolution Convolutional Neural Network Algorithm in Image Optimization for High Speed Photography

针对高速摄影系统实时采集的图像几次局部放大后的优化问题,对超分辨率卷积神经网络算法(SRCNN)的非线性映射层进行调优,增大滤波器f2值,并与传统的双三次插值算法仿真处理后的结果进行比较。仿真结果表明:调优后的SRCNN算法具有更高的平均峰值信噪比PSNR。

The optimization problem of the image after several times of partial enlargement was studied in this paper.The nonlinear mapping layer of the super-resolution deep learning algorithm Super-Resolution Convolutional Neural Network(SRCNN) was optimized and the size of the filter f 2 was increased.Then the processed result was compared with that by the traditional bicubic interpolation algorithm.The simulation results showed that the optimized SRCNN algorithm had higher peak signal-to-noise ratio(PSNR).