Deep learning-based image defogging algorithm

Abstract

The invention discloses a deep learning-based image defogging algorithm, and the algorithm is used for removing fog interference in foggy images so as to reduce the influence of fog on the image quality. The algorithm comprises the following steps of: 1, obtaining a training sample set and a test sample set; 2, carrying out HSL space variation on foggy images in the sample sets, extracting local low-brightness characteristics of the foggy images, and carrying out scale zooming and normalization processing on all characteristic components; 3, finding out discrimination perspective ratio so as to enable a depth discrimination neural network to realize adversarial training; 4, training the abovementioned characteristic components by utilizing a depth generation adversarial neural network, and learning to establish a mapping network between the foggy images and perspective ratios; and 5, carrying out a defogging test on the test sample set by utilizing the depth generation neural network. According to the defogging algorithm disclosed by the invention, a mapping relationship between the foggy images and the perspective ratios is established through a deep learning algorithm, so that the problem that a previous defogging algorithm is lack of prior information is effectively solved, and a relatively good defogging effect is achieved.

Description

technical field [0001] The invention relates to the fields of image processing technology, pattern recognition and artificial intelligence, in particular to an image defogging algorithm based on deep learning. Background technique [0002] In recent years, with the continuous development of artificial intelligence, more and more image acquisition devices have the ability to automatically identify targets in different scenarios. However, this automatic recognition ability is often restricted by environmental factors in the scene, especially the existence of fog or haze, which reduces the saturation and contrast of the target in the image, which has become an important factor affecting the automatic recognition effect of the device. Therefore, a Effective and feasible dehazing algorithm has important theoretical and practical significance. [0003] Traditional dehazing algorithms are mainly divided into two categories: dehazing algorithms based on atmospheric light scattering...

AI Technical Summary

Problems solved by technology

For the existing defogging methods, due to the lack of effective prior knowledge in foggy images, the optimal perspective rate cannot be obtained, and the phenomenon of color shift appears in the process of image restoration.

Images