Method for converting thermal infrared image into visible-light-like color image based on CycleGAN

Abstract

The invention discloses a method for converting a thermal infrared image into a visible-light-like color image based on a CycleGAN, and belongs to the field of image processing based on a generative adversarial network. The implementation method comprises the following steps of: according to image domain conversion information of a thermal infrared image and a visible light image in a database, connecting two generators for converting the thermal infrared image into a color visible light image and converting the color visible light image into the thermal infrared image by utilizing the cyclic consistency loss of a cyclic generative adversarial network CycleGAN; high-precision pairing of the thermal infrared image and the color visible light image is not needed, and the purpose of unsupervised learning from the thermal infrared image to the color-like visible light image is achieved. And a full convolutional neural network structure is further adopted in the network of the generator, so that the size of the input thermal infrared image is not limited. According to the learning of the image domain conversion information of the thermal infrared image and the visible light image in the database, the natural color rendering of the thermal infrared image is realized, and the observation adaptability and the target identification and detection speed are improved.

Description

technical field [0001] The invention relates to a method for directly converting grayscale images of day and night thermal imaging into color images (or natural color images) of the same scene similar to visible light, and belongs to the field of image processing based on generative confrontation networks. Background technique [0002] With the development of infrared focal plane detector IRFPA technology, thermal imaging technology is widely used in military and civilian fields. It is less affected by environmental lighting and weather changes, and can provide people with images of target scenes even at night. . However, the radiance distribution of the target scene reflected by the infrared thermal imaging image is very different from the visible light brightness image distribution of the human visual habit. Long-term observation of the human eye is not only prone to fatigue, but also directly affects the perception of the target (speed and high-level ability) and context...

AI Technical Summary

Problems solved by technology

One of the factors is the use of two separate detectors, which not only affects system miniaturization (volume, weight, and power consumption), but also brings major problems to system configuration and image processing.

As a result, multi-color / two-color focal plane detectors have been developed, and the detector technology has become more complex. At present, two-color focal plane detectors are mainly introduced, and the dual-band fusion method is not yet mature, and it is difficult to obtain satisfactory color fusion image effects.

Images