Research On Image Restoration Technology Under Non-uniform Illumination

In low illumination and complex night environment,the complexity of scene will degrade the image quality,such as uneven illumination,few details and lack of edge information,which will affect the accuracy of intelligent driving assistance system in lane recognition,vehicle identification and pedestrian recognition.Therefore,image restoration processing is required.For images under non-uniform illumination,the classical image restoration methods include grayscale transformation,homomorphic filtering,Retinex restoration and exposure compensation.However,these methods only consider the illumination effect of the light source in the two-dimensional image,but do not consider the spatial relationship between the light source and the illuminated object.In addition,for the entire image,the same parameter is used for adjustment,so in the urban night scene with complex illumination,so it is easy to cause problems such as overexposure or underexposure in urban night scenes with complicated illumination.Aiming at the above problems,this paper proposes an image restoration model based on spatial relationship,and implements an image restoration system based on adaptive exposure compensation technology.This system can recover urban nighttime non-uniform illumination images under complex illumination,avoiding problems such as overexposure or underexposure.Before image restoration,this paper first uses mapping method to detect multiple light sources in the image,then uses optical flow method based on Deep Learning(DL)to predict the direction of light source movements,and to determine the pixel distance diference between the light source movements in the two frames.This paper presents a measurement method based on a single camera,combining with the output of optical flow method,a model based on Parallax Theory(PT)is used to calculate the distance "F" between the light source and the camera.The mapping relationship of the two-dimensional image to the three-dimensional space is calculated by Zhang plane calibration method.In this way,the physical distance "f" of the light source to the camera in the actual scene is measured.In image restoration stage,according to the spatial relationship between the light source,the illuminated object and the camera,the image restoration model can be established,of which formula is(?).And the distance "h" from the light source to the illuminated object can be obtained.In addition,combined with reflection coefficient illumination intensity,light attenuation coefficient and other factors,the adaptive exposure compensation coefficient "?(h)" can be obtained according to the formula ?(h)=r×I0× exp(-khc).Finally the image can be restored by the adaptive exposure compensation algorithm.In order to compare the effects of reconstructing images based on adaptive exposure compensation and classical methods,this paper have used non-uniform illumination images of urban nights under complex illumination for testing.The experimental results show that compared with the classical image restoration method,the PQM(Perceptual Quality Measure)value of the restored image obtained by the adaptive exposure compensation restoration system is closer to the ideal value of 10,M(Image Mean Gray Values)and F(Contrast Enhancement Factor)values of images are more reasonable.