Unsupervised image defogging method based on multi-scale depth image prior

Abstract

The invention relates to an unsupervised image defogging method based on multi-scale depth image prior, and belongs to the technical field of computer vision images. The method comprises the following steps of: 1, down-sampling an original image to generate small-size image prior; firstly, respectively inputting three noise images with the same size as a downsampled foggy image into three neural networks with codec structures to obtain three intermediate results representing an atmosphere illumination image, a transmission image and a defogged clear image; and then modeling the three intermediate results by using an atmospheric scattering model to obtain a reconstructed foggy image. and 2, inputting a noise image with the same size as the original size image into the same network, and initializing the network by using the prior obtained by the small-size image. The method is reasonable in design, fully considers the problem that the prior extraction of an unsupervised defogged image is difficult, reduces the prior extraction difficulty through a multi-scale method, and improves the visual effect and stability of a reconstructed image.

Description

Technical field [0001] The present invention belongs to the field of computer visual image, in particular, a non-supervised image densification method based on multi-scale depth image prior. Background technique [0002] Smog is a typical atmosphere, which is accumulated from small water droplets, dust, smoke or other particles to accumulate in the air. These particles will absorb and scatter light, directly leading to a decrease in visibility. The image contrast taken under such weather conditions will be deteriorated, lose visual details, which will bring difficulties for subsequent applications. In addition to the effects of direct visual effects, ensure image quality or images, high-level visual applications, such as target detection, semantic splitting, etc. Therefore, the image de-fog is an extensive study as an image pretreatment and visual enhancement technique in recent years, and has achieved more significant results. [0003] Image Dehazing is processed for problems su...

AI Technical Summary

Problems solved by technology

However, like neural networks in other tasks, these deep learning-based image defogging methods rely on a large-scale training data set, and the introduction of the data set will inevitably bring about the problem of data set pairing and image domain coverage.

Images