Image defogging method based on deep convolutional neural network under Bayesian framework

Abstract

The invention relates to an image defogging method based on a deep convolutional neural network under a Bayesian framework, and belongs to the technical field of image processing. The method comprises the following steps: firstly, obtaining a synthetic foggy day image data set ITS as a training set, and completing Bayesian model modeling for the training set; secondly, obtaining a function for optimizing network parameters through formula transformation; and finally, inputting the synthesized foggy day image data set into a deep convolutional neural network model (BDCNet), in the training process, calculating a loss rate through a new loss function, continuously iterating and updating network parameters, obtaining an optimal defogging model, and carrying out image defogging operation. The model provided by the invention does not depend on an atmospheric scattering model any more, image features can be directly learned through the model, and image details can be better recovered in combination with prior knowledge. The defogging result of the model provided by the invention is better in visual effect, and the applicable scene range is wider.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an image defogging method based on a deep convolutional neural network under a Bayesian framework. Background technique [0002] In smoggy weather, the fine dust and water vapor floating in the atmosphere will not only endanger human health, but also reduce the visibility of roads in smoggy weather, and the images captured by the machine will also have problems such as decreased contrast, unclear details, and dim colors. These low-quality pictures and videos have a bad impact on advanced machine vision systems, such as traffic monitoring systems, driverless systems, object detection systems, joystick systems, and more. Therefore, image dehazing plays a vital role in improving the recognition ability of subsequent advanced vision systems. [0003] At present, there have been many studies on image defogging algorithms at home and abroad. Image defogging algorithms can be divided in...

AI Technical Summary

Problems solved by technology

The model works well for high-resolution images, however, the model uses a deconvolution method, and the dehazed image will appear grid-shaped noise

The image dehazing method based on deep learning is the most widely studied, but its dehazing performance is still affected by the estimation of transmission map and atmospheric light and the structure of deep neural network

Therefore, the dehazing method based on deep learning has certain limitations and needs to be further improved and improved.

Images