Research On Low-Light Image Enhancement Algorithm Based On Neural Network And Attention Mechanism

With the continuous development of photography technology,the resolution of images has been significantly improved.However,in the process of acquisition and shooting,problems such as insufficient brightness,color distortion,loss of details and low signal-to-noise ratio are often caused by many uncontrollable factors such as indoors,nighttime,or cloudy days.To deal with these problems,Low Light Image Enhancement(LLIE)technology came into being,which aims to generate normal light images with high quality,suitable brightness,and clear details,to improve the visual effect and usability of images.This thesis will study low-light image enhancement tasks from the following aspects based on neural network and attention mechanism methods.(1)To solve the problem that existing algorithms are difficult to process brightness information and color information at the same time,resulting in color distortion in image enhancement results,this thesis proposes a low-light image enhancement algorithm that combines U-Net and color correction models.Aiming at the problem of color deviation in low-contrast,low-signal-to-noise ratio and low-light scenes,a strategy of converting color channels is adopted to decompose the original RGB image into brightness and color components.The luminance component is then enhanced by the U-Net network,while the color component is improved by the Color Correction Network(CC-Net).Finally,the enhanced luminance and color components are converted back to RGB color space to obtain the enhanced image.The experimental results show that the algorithm makes full use of the brightness and color information of the image,effectively improves the brightness of the image,reduces color distortion,and is superior to the comparison algorithms in terms of visual effects and objective evaluation indicators.(2)To solve the problem that existing algorithms are difficult to adaptively process unevenly distributed brightness and noise,resulting in uneven exposure of enhanced results,this thesis proposes a low-light image enhancement algorithm based on LAB color space and multi-scale feature pyramid.The algorithm divides the entire enhancement process into two steps.First,the input image is converted from the RGB color space to the LAB color space,and the illumination attention map is calculated through the brightness component of the L channel,and the receptive field is further expanded by dilated convolution to generate noise attention.try hard.Two attention maps are then utilized to adaptively enhance brightness and remove noise in a multi-scale feature pyramid network.Experimental results show that the algorithm can effectively improve the problem of uneven exposure in low-light images and reduce unnecessary noise.The comparative experiments on different data sets also show that the enhancement effect of the algorithm in this thesis is more suitable for the visual perception of the image by the human eye.