A Method for Dehazing Complicated Scene Images Based on Semi-Trained Generator

Abstract

A complex scene image defogging method based on a semi-trained generator belongs to the field of image defogging applied to complex environments, including a training process and a use process. During the training process, first, use the CycleGAN network to train on any image defogging data set, output a real-time defogging image every 50 training times and save the current model, and the training ends after 2000 times; secondly, repeat the above process ten times; finally , select the haze-free grayscale image with the best dehazing effect and no color information in the saved rendering image, and use its corresponding generator G in the saved model as the final sketch module. When in use, input the foggy image of any scene into the sketch module, and the grayscale image after dehazing can be output. The scope of application of the present invention is not limited to training data sets, the method has strong adaptability, visibility and authenticity, can be applied to any scene, and can help intelligent systems to be able to play a certain role.

Description

technical field [0001] The invention belongs to the field of image defogging applied in complex environments, and relates to an image conversion-like single image defogging method based on a semi-training generator. The image defogging technology can help an intelligent system to perform normally under foggy weather conditions. However, the current image defogging technology is rarely able to deal with a variety of scenes, and can only be applied to several fixed environments. Background technique [0002] In the field of computer vision, the quality of images has a great influence on the completion of tasks such as target detection and image recognition. However, images acquired in real application environments are often affected by suspended matter in the air (fog, haze, dust, etc.). These effects can blur the image and eventually make it difficult to extract effective features of the image. Research on image defogging technology can restore the image affected by fog to ...

AI Technical Summary

Problems solved by technology

[0005] However, these methods enhance the effect of dehazing by enhancing the general performance of CycleGAN. There is no research on the characteristics of fog. The characteristics of fog in some images are difficult to distinguish. The model’s analysis of the image is not sufficient enough to completely separate the fog from the image In the process of image conversion, it is interfered by the residual fog information, and the effect of defogging is not particularly ideal. Moreover, it is still necessary to train the corresponding scene in order to perform a good image defogging work.

Images