Real-time image dehazing method based on gpu

Abstract

The present invention relates to a real-time dehazing method for images based on GPU, comprising the following steps: (a) acquiring original images and atmospheric light values; (b) allocating a first shared memory, and using the first shared memory according to the atmospheric light values Calculate the first transmittance map; (c) allocate the second shared memory, the third shared memory, the fourth shared memory and the fifth shared memory, and use the second shared memory, the third shared memory, the The fourth shared memory and the fifth shared memory perform two guided filtering according to the original image and the first transmittance map to obtain a haze-free image. In the embodiment of the present invention, by implementing the method of secondary guided filtering on the GPU, the finally obtained dehazing image is more ideal.

Description

technical field [0001] The invention belongs to the field of visual image processing, and in particular relates to a real-time image defogging method based on GPU. Background technique [0002] Smog has been a popular word in society in recent decades, and there are not a few countries affected by smog in China and even in the world. Fog and haze lead to unclear vision and sometimes have extremely bad effects, such as foggy driving, foggy drills, foggy target recognition, and so on. Therefore, it is very important to study the dehazing algorithm with stable dehazing effect and good real-time performance. [0003] However, there is no dehazing algorithm that can be really applied in real life. The reason is that the algorithm with stable dehazing effect has a large amount of calculation and does not have real-time application prospects, while some current real-time dehazing algorithms are There is nothing to do with video images, there are often unnatural transitions betwee...

AI Technical Summary

Problems solved by technology

It can be seen that in the final defogged image, the defogged effect is not ideal, such as the tree branches in the upper right corner, the pillars next to the railroad tracks, etc., mainly because the edge information of these areas in the haze image itself is not clear enough, so the use of fog Filtering the haze image as a guide map is not ideal for these areas

Moreover, the performance of these two algorithms on the CPU is not satisfactory. When processing large images (such as standard 1080p images), these two algorithms do not have the ability to process standard high-definition images in real time on the CPU. The real-time performance on embedded devices is even worse, and it is basically impossible to apply it to miniaturized devices

Images