Image rain removal method based on multi-density rain streak perception

Abstract

The invention discloses an image rain removal method based on multi-density rain streak perception, and the method comprises the steps: obtaining an original image O with rain and a low-frequency auxiliary image IRG, constructing a low-frequency rain-free image extraction network, and obtaining a low-frequency part B1 of a target rain-free image; constructing a high-frequency multi-scale rain density sensing network to obtain high-frequency part mixed information psi(Bh, Si) of the original image O with rain; and fusing the original image O with rain, the low-frequency part B1 of the image O with rain and the high-frequency part mixed information psi(Bh, Si) of the original image O with rain, and constructing a rain removal sub-network to obtain a target rain-free image. According to the method, various rain type scenes in real life can be effectively processed, high-quality rain-free clear images can be generated under different rain densities, and the method has high environmental adaptability.

Description

technical field [0001] The invention belongs to the technical field of computer vision, in particular to an image deraining method based on multi-density rain streak perception. Background technique [0002] With the development of computer vision and image processing technology, various imaging systems are widely used in video surveillance, automatic driving, remote sensing observation and even military fields. However, the imaging system is easily affected by bad weather outdoors, such as rain and snow, resulting in reduced contrast, blurring, and loss of detail information in the acquired images, which greatly reduces the accuracy and reliability of subsequent high-level visual task analysis and processing. . Therefore, it has important practical significance and wide application value to eliminate the influence of rainy and snowy weather on the image scene and to clear the degraded image in rainy day. [0003] Image deraining has been an important research topic in the...

AI Technical Summary

Problems solved by technology

At present, there are four main types of rainwater removal methods with relatively ideal effects: (1) Rain removal models based on purely physical models and mathematical derivations. However, due to the complexity of the formation mechanism of rain and fog in reality, it is difficult for simplified physical models to achieve better adaptation. sex

(2) The rain removal method based on image processing knowledge, its representative work is similar to the principle of image defogging, using the method of dark channel prior to image rain removal, the disadvantage of this type of method is that the model is relatively simple, and the transmittance estimation Too rough, easy to lose details

(3) A rain removal method based on sparse coding dictionary learning and classifiers. This method assumes that the characteristics of rain and background are separable. By optimizing the dictionary and coding, the rainy picture is divided into two codes of the same dictionary. and, these two encodings represent the encoding of the rain and the encoding of the background, respectively. The disadvantage is that such methods are prone to failure when the rain streaks and the background have very similar shapes

However, the method of distinguishing high and low frequencies of rain images proposed by them is still based on the above physical model, and has made strong assumptions and restrictions on the parameters of the high and low frequency parts of the physical model, which cannot fully represent the complexity between background information and rain streaks in rain images. Coupling relationship

In addition, most of the existing rain removal methods are limited to a specific rain model, which cannot adapt to the application scenarios of various rain types

Images