Super Resolution Reconstruction Algorithm of Single Frame Infrared Image Based on ESRGCNN
Super-resolution reconstruction algorithm of infrared image is the research focus in the field of image processing algorithm in recent years.The existing convolutional neural networks(CNNs)with strong learning ability will improve the effect of image super-resolu-tion reconstruction while increasing the computational cost,and the subsequently proposed enhanced super-resolution group convolutional neural network(ESRGCNN)with shallow structure not only saves cost but also has high efficiency in the super-resolution reconstruction of visible images.Therefore,in view of the shortcomings such as poor resolution and low contrast of infrared images,the final super resolution infrared image is obtained by weight construction of the high resolution texture detail image obtained from the preprocessed infrared image through high-frequency texture detail extraction,reconstruction and other operations,and the high-frequency detail layer and base layer of the infrared image obtained through ESRGCNN network,and weight fusion after CLAHE processing.A large number of comparative experi-ments on the infrared dataset CVC-14 show that the PSNR of the optimized algorithm proposed is about 13.7%-32.4% better than that of the classical algorithm in three kinds of magnification reconstruction images,and the SSIM of its reconstruction effect is about 13.9%-32.4% better than that of the classical algorithm.
国家科技期刊平台
NSTL
万方数据
