Image defogging method based on deep learning

Abstract

The invention discloses an image defogging method based on deep learning, and the method comprises the steps of building an image defogging model based on deep learning, forming a defogging model considering an atmospheric light value and transmissivity, carrying out the preprocessing of a to-be-processed image, and forming an image with a unified size; performing DS convolution processing and pooling processing on the preprocessed to-be-processed image in sequence, then obtaining a transmittance value through up-sampling processing, thus improving the calculation speed; performing expansion convolution processing on the preprocessed to-be-processed image in an expansion convolution layer, performing pooling processing on the image after expansion convolution processing, and subjecting the pooled image to dimension reduction to obtain images with the sizes of 256, 10 and 1, namely, an atmospheric light value can be obtained; substituting the obtained transmissivity value and the atmospheric light value sequentially into an image defogging model, and obtaining a fogless image. According to the invention, more original image information can be reserved, the defogging effect is clean, and high robustness is achieved.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an image defogging method based on deep learning. Background technique [0002] Hazy weather is a common weather phenomenon, and the particles contained in the hazy weather cause atmospheric light to scatter, resulting in a serious reduction in the quality of captured images. In the case of smoggy weather accompanied by sand and dust, it will even seriously interfere with people's visual problems. Therefore, image processing for dehazing has an important positive effect on image research and social development, and many excellent dehazing algorithms have been born. Currently, image enhancement and image restoration are the two main types of dehazing algorithms. The image enhancement algorithm does not consider the essential cause of image degradation, and only starts with increasing image contrast and highlighting useful details. Although it has the effect ...

AI Technical Summary

Problems solved by technology

The image enhancement algorithm does not consider the essential cause of image degradation, and only starts with increasing image contrast and highlighting useful details. Although it has the effect of defogging to a certain extent, it loses a lot of image details. For foggy vehicle detection images, its own Due to the influence of light and impurities, the image details are not obvious enough, and the vehicle features are not prominent enough, so the image enhancement algorithm is not used

Based on the physical model is a widely used method in the current dehazing algorithm research. In 2011, He et al. proposed a theoretical hypothesis of dehazing prior to the dark channel color, using guided filtering to refine the transmittance to restore the image. When the brightness of the object is close to the brightness of the sky, this prior defogging method based on statistical laws will not be able to achieve a good defogging effect; Zhu et al. proposed a color attenuation prior defogging algorithm, by establishing a foggy The linear model of the image and the use of supervised learning methods to restore the depth information of the scene, this method can restore more detailed information, but there is still a certain degree of residual fog; Wang et al. proposed a dehazing algorithm based on linear transmission, the The algorithm runs faster, but there is a phenomenon that the more thorough the defogging is, the darker the overall restored image will be

Images