Infrared image super-resolution reconstruction method based on generative adversarial network

Abstract

The invention discloses an infrared image super-resolution reconstruction method based on a generative adversarial network, and belongs to the field of computer vision. According to the method, two aspects of a generation network and a loss function of an existing algorithm SRGAN are improved; in the improvement of the structure of the generation network, the generation network is combined with atraditional bicubic interpolation method; in the improvement of a loss function, in order to obtain high objective evaluation indexes (peak signal-to-noise ratio and structural similarity) while a good visual effect is achieved, pixel-by-pixel mean square error loss is added into the loss function of a generation network. Compared with the original SRGAN algorithm, the improved algorithm has the advantages that the low-frequency region of the reconstructed image is smoother, artifacts are reduced, high-frequency details are clearer, and the peak signal-to-noise ratio (PSNR) and the structuralsimilarity (SSIM) of the objective evaluation index are both improved.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to an infrared image super-resolution reconstruction technology based on a generation confrontation network capable of obtaining clear high-frequency details. Background technique [0002] Infrared imaging refers to a technology that uses a detector to receive infrared rays reflected by a target object and obtain an infrared image through photoelectric conversion. Infrared imaging technology has many applications in civil and military fields. Infrared imaging technology has the advantages of strong ability to penetrate smog and dust, long detection distance, little influence by external light, and sensitivity to thermal radiation of objects. With these advantages, infrared imaging technology can be widely used in the fields of quality inspection and safety inspection, and can also assist visible light imaging in environments with smog or poor lighting conditions, and work...

AI Technical Summary

Problems solved by technology

But its disadvantage is that there will be artifacts in the low-frequency area, that is, the texture that does not exist, and the details of the high-frequency area will be blurred after zooming in. The objective evaluation indicators of the generated image, the peak signal-to-noise ratio PSNR and the structural similarity SSIM are too low, which does not match the subjective visual impression.

Images