Night image defogging algorithm based on deep learning

Abstract

The invention discloses a night image defogging algorithm based on deep learning. The end-to-end network structure is more suitable for engineering practicability. The characteristics of non-uniform illumination, serious color cast, low brightness and the like of night images exist, a traditional defogging algorithm based on an atmospheric scattering model is not applicable, the method focuses on the powerful nonlinear fitting capability of deep learning, a U-Net network is combined with dual-tree complex wavelet transform, and the defogging efficiency is improved. More residual structures are fused into a network model to extract the structure and texture information of a night foggy image, and an attention module is introduced into the network to effectively quantify the relative importance between feature layers. Through a self-constructed foggy-fogless night data set, an error is calculated according to a global-local loss function, network weight is adjusted based on error back propagation, and a completely trained night image defogging network model is finally obtained through multiple iteration training. The method provided by the invention can restrain the mesh artifact problem caused by frequent up-sampling and down-sampling in the defogging task.

Description

[0001] The invention belongs to the field of digital image processing, in particular to a night image dehazing algorithm based on deep learning. Background technique [0002] With the increase of haze weather in recent years, the resolution of the collected images has decreased significantly, especially in the low-visibility environment at night. Compared with imaging in a foggy environment during the day, due to the low visibility at night, the image captured by the observation is not clear, which is reflected in the image data, and the pixels with high gray value in the image are weakened. In addition, the color and contrast of the scene change significantly compared to the daytime, and the characteristic information contained in it is weakened, which greatly reduces the recognizability of the scene target. If accompanied by severe weather such as haze, the transparency of the atmosphere will be further reduced, and the impact on high-resolution observation activities will b...

AI Technical Summary

Problems solved by technology

These two methods have better defogging effects on images in daytime environments. Since the sky and atmospheric light are no longer the only source of light energy in the scene in the night imaging environment, a large number of artificial light sources participate in imaging with different scattering paths. Fog technology methods are difficult to directly apply to the restoration of foggy images in night environments

Images