Research On Low Light Image Enhancement Algorithm Based On Convolutional Neural Network

With the popularity of professional digital imaging equipment,digital image processing is widely used in industrial production,video surveillance,intelligent transportation,remote sensing and monitoring and many other fields and play an important role in it.However,due to the image acquisition system in the image acquisition process,due to the influence of various uncontrollable factors,especially in indoor lighting,night lighting,cloudy sky and other unfavorable conditions,the image acquired by the image acquisition system often has a relatively low contrast ratio,low dynamic range intensity,the image dark area and light area details disappear and other defects.Therefore,how to obtain clear still or moving images under light conditions has become a problem to be solved.For this reason,image enhancement technology has attracted widespread attention and attention from industry and academia because image enhancement not only meets the need for visual experience,but also improves the reliability and robustness of outdoor visual systems,making it easier for image processing systems to analyze and process images.In order to address the problem of difficult image recognition under low-light conditions that occur daily,the following methods are proposed to study the enhancement of dark-light images.1.In response to the over-enhancement or unnatural effects,artifacts and other defects of the current low-light image enhancement algorithm,this paper proposes a CNN-based low-light image enhancement method combining the characteristics of color model transformation algorithm and convolutional neural network.Unlike the original conventional method,this method does not require estimation of the illumination image and reflectance image and generates enhanced images directly endto-end.Using the color model transformation algorithm,the luminance component I is extracted separately from the input image and the luminance component I is enhanced by CNN to obtain a new illuminance component I? to obtain the enhanced image.2.For low-light images have low contrast,and common image enhancement tools have problems such as slow speed,bright areas may be over-enhanced,and details are lost.In this paper,we propose a contrast enhancement method for low-light images based on the Sigmoid function,which is based on the YCbCr color space to enhance the contrast of low-light images without affecting the input image color information,and the obtained images have good enhancement effect.3.In response to the shortcomings of the classical Retinex algorithm for image enhancement,such as image coloration,a method for enhancing low-light images in combination with CNN and Retinex is presented.Using the low-light image as input,CNN learning is used to predict the mapping relationship between the low-light image and the corresponding luminance image and output its luminance map,then the estimated luminance map is optimized by Gamma correction adjustment,and the resulting luminance map is combined with the classic Retinex model to enhance the low-light image.In contrast to the experimentally validated and compared algorithms,the algorithm proposed in this paper can improve the overall brightness and contrast of the image while reducing the effects of uneven lighting and improving the image quality and clarity.