| Vision is an important way of promoting human access to information.Image carries a lot of information that people need in daily life,work and research.Night,as an unavoidable scene with everpresent environment with poor lighting conditions,seriously affects quality of the images,usually resulting in low brightness and visibility of detail,high level of noise and occur color deviation.This kind of quality degradation affects subsequent use of images.In order to improve the perception quality of nighttime image,this paper studies nighttime image enhancement methods to achieve the main goal to enhance the brightness,suppress noise amplification and correct color deviation.Based on the specific characteristics of nighttime images,this paper proposes solutions to enhance the quality of various kinds of images taken in night scene.Through in-depth study of the theory and technology of image quality enhancement,the advantages and disadvantages of various methods in practical application are analyzed.And several nighttime image enhancement methods are further proposed.Based on the statistical characteristics and prior knowledge of natural images,several nighttime image enhancement methods are proposed.The research results and innovations are summarized as follows:1.Be directed against the low brightness and visibility of nighttime image,a nighttime image enhancement method based on improved Retinex model is proposed.This paper finds that the inverted night image and the dense fog image have very similar histogram distribution.Therefore,introduce haze removal method into this task and prove the theoretical consistency between this mehtod and Retinex.According to the fact that nighttime images always contain darker regions,the value of atmospheric illumination is set to the highest brightness value.A multi-scale edgepreservation median filtering method is designed to improve Retinex model and simplify the atmospheric scattering illumination model.Eventually,realize the realtime enhancement of nighttime images.The experimental results show that the proposed method can improve the visual perception quality of nighttime images,and effectively suppress the over-enhancement phenomena such as loss of detail and brightness distortion.2.In order to suppress the noise amplification caused by the enhancement of image brightness,a nighttime image enhancement method based on multi-layer image decomposition is proposed.This paper regards image as the sum of three layers: structure layer,texture layer and noise layer.The structure layer includes the contour of the object and the main edge information.The texture layer contains texture details and smaller edges.Firstly,the structure layer of the image is separated,and then enhanced by the adaptive brightness improvement method combining global illumination and local illumination.Then,the global illumination map is used to calculate the mask layer to suppress the noise intensity.A weighted least squares algorithm is used to separate the noise layer and obtain a smooth texture layer.Finally,combine the enhanced structure layer and the smoothed texture layer to obtain the result.The experimental results show that the proposed method can effectively enhance the brightness of nighttime images and suppress noise.3.Considering the low brightness,complex noise and color deviation of nighttime images,in view of the fact that the nighttime image with color deviation does not meet the gray world assumption,a nighttime image enhancement method based on weighted Total Nuclear Variation(TNV)is proposed.In this paper,an adaptive initial illumination map calculation method is proposed based on analysis of the histogram statistics of nighttime images.The illumination map is smoothed using TNV model.Then a weight matrix is calculated using the illumination map to assist TNV model suppress the noise in the reflection map.Finally,the brightness transformation model and the camera response model are used to enhance the input nighttime image.The experimental results show that,compared with the existing image enhancement algorithms,the proposed nighttime image enhancement method can achieve better enhancement effect,suppress noise and correct color effectively.4.For the problem of hard improving the perception quality of extremely low illuminance imagesa and cannot recovering the missing details by traditional image processing methods,a nighttime image enhancement method based on Dense-net is proposed.This paper collects 1100 high-quality natural images from open data-sets to generate low-illumination images via various brightness transformation methods to construct synthetic low illumination image data set,and adds the same scene of High Dynamic Range(HDR)image to enrich the dataset.The proposed nighttime image enhancement based on Dense-net can solve the problem of brightness enhancement of extremely low illumination image.The experimental results show that the proposed method can enhance nighttime image including extremely low illumination image,and keep the details and structure information of the input images. |
