Physical model-based low-illuminance image enhancement algorithm

Abstract

The invention discloses a physical model-based low-illuminance image enhancement algorithm. The physical model-based low-illuminance image enhancement algorithm is a Retinex algorithm and dark channel prior theory-combined improved algorithm. The objective of the invention is to solve the problem of little possibility of achieving balance among brightness improvement, contrast improvement and noise removal of an existing low-illuminance image enhancement algorithm. According to the physical model-based low-illuminance image enhancement algorithm, a BM3D algorithm is improved in efficiency, and the improved BM3D algorithm is applied to a YCbCr space; a brightness spread map is coarsely estimated in an HSI space; an atmospheric physical model under a low-illuminance condition is improved; and the Retinex algorithm is used in combination so as to refine the brightness spread map. As indicated by an experiment, the physical model-based low-illuminance image enhancement algorithm can achieve comprehensive enhancement of the brightness and contrast of a low-illuminance image and is also significantly improved in computing speed.

Description

technical field [0001] The invention relates to an image enhancement algorithm, in particular to a low-illuminance image enhancement algorithm based on a physical model. Background technique [0002] In low-light conditions, such as rainy days, nights, or in mines, if there is no auxiliary light, the image quality is often poor when collecting images. These images not only look poor to the human eye, and many key elements are difficult to see clearly, but also bring great difficulties to machine recognition and monitoring and tracking, which greatly reduces the practical applicability of low-light images. In order to improve the quality of low-light images, low-light images with poor image quality must be enhanced and denoised. Traditional image enhancement methods do not consider the characteristics of low-light images, and the effect is not ideal. In recent years, research on algorithms for low-illuminance image enhancement has also made great progress. These methods can...

AI Technical Summary

Problems solved by technology

[0002] In low light conditions, such as rainy days, nights, or in mines, if there is no auxiliary light, the image quality is often poor when collecting images

These images not only look poor to the human eye, and many key elements are difficult to see clearly, but also bring great difficulties to machine recognition and monitoring and tracking, which greatly reduces the practical applicability of low-light images. In order to improve the quality of low-light images, it is necessary to enhance and denoise the low-light images.

Traditional image enhancement methods do not consider the characteristics of low-light images, and the effect is not ideal

In recent years, research on algorithms for low-illuminance image enhancement has also made great progress. These methods can be divided into the following categories: one is based on the improvement of the histogram equalization algorithm. The information is concentrated at the lower end of the brightness, and the histogram equalization algorithm merges the gray level and expands the gray distribution range, which makes the processing results of this type of algorithm often unsatisfactory; the algorithm based on Retinex has received the most extensive attention , but the large amount of calculation and the unavoidable halo effect limit its application, and even the image with extremely low illumination will be grayed out; the method of multi-image fusion requires high-quality image information in the same scene. In addition, there is another type of algorithm for color space enhancement, which is generally limited because it is not practical for extremely low-light images, and the calculation is very large

[0003] In 2011, scholars from Tsinghua University proposed a low-illuminance image enhancement algorithm based on defogging technology. The low-illuminance image is enhanced by applying the defogging method after inverting the low-illumination image. Spots will appear in continuous places, and the details are not fine enough

Images