Super-resolution of infrared image
Yudan Chen, Jie Liu, Ming-quan YANG, Gang Li · Eighth Symposium on Novel Photoelectronic Detection Technology and Applications · 2022
The super-resolution of infrared images was discussed. Three classical super-resolution algorithms as Bicubic Interpolation, Projection Onto Convex Sets(POCS) and convolutional neural network(SRCNN) were applied. The experimental results showed that the three algorithms on infrared images were not as good as on visible images due to the imaging differences between infrared image and visible image. According to the results of PSNR(peak signal to noise ratio) and SSIM(structure similarity image measure) index parameters, the reconstruction effects of SRCNN algorithm were better than the bicubic interpolation and the projection on to converge sets (POCS) algorithm. The super-resolution effect of model 2 using infrared image database as training sample was better than that of model 1 using visible image database as training sample. It can be deduced that the effect of super-resolution of infrared image based on convolutional neural network can be improved using infrared images as training sample database.