The Image Quality Improvement Technology For Visible Night Scene

Visible night images might suffer from noise,blur,limited HDR and over exposure due to their weak illumination if not photographed appropriately.These issues heavily effect the accuracy of collecting and processing optical information.The research of image quality improvement technology for visible night scene has important practical significance for reducing the hardware requirements of the imaging system and improving its photographic performance to meet users' wider production and life needsThe long exposure requirement for visible night images results in blur.Large ISO requirement results in large noise.The dynamic range requirement for night view results in insufficient dynamic range and over exposure.In response to the above comphrehensive problems,the research improves image quality from three aspects:denoise,deblur and over exposure restorationOn the issue of noise in visible night images,this paper focuses on the spatial correlation of noise.The spatial independent noise signal turns to spatial related because of color interpolation and color space transformation operations in ISP.Based on the noise probability model and the image processing process,this paper proposes a noise calibration model based on hardware calibration to solve its spatial correlation problem Experiments show that the modified algorithm can effectively solve the residual clump noise in the original algorithm,suppress ringing,and improve resolutionOn the issue of overexposure of visible night scene images,this paper proposes a modified haze model to describe the phenomenon of overexposure.This model assumes that the overexposed area is a non-uniform haze layer caused by a light source or a highly reflective surface in a normal exposure image.Hence,this paper proposes an overexposure correction method based on haze removal model and image fusion technology.Experiments show that compared with the traditional overexposure restoration algorithm or HDR correction algorithm,the restoration result of this algorithm has more details,less noise and more realistic color informationOn the issue of blur in visible night images,this paper first theoretically derives and experimentally verifies the technical route of blur kernel estimation using attitude sensor sampling.Hence,this paper proposes a blind deblurring method based on inertial sensor and Short-Long-Short exposure strategy.The core idea of the algorithm is to collect the motion state of the imaging system during the exposure process with the inertial attitude sensor information,and correct the sensor's data error by using Short-Long-Short exposure strategy.to obtain a more accurate blur kernel estimation.For the estimated blur kernel,we gradually refine the estimated blur kernel through the iterative process of semi-blind restoration to further improve its accuracy.Comparing with the other rule-based state-of-the-art deblurring methods,the proposed method has faster computational speed,wider scene adaptability and better noise and ringing suppression.Overexposed pixels in night blurred images can cause serious interference in deblurring.This paper proposes an edge-aware scales recurrent network,i.e.,EASRN.This method solves the interference from overexposed pixels by constructing a more realistic dataset.In the term of network architecture,this method employs a scale recurrent network with separated deblurring subnet and upsampling subnet.In the term of loss function,this method adapts a novel edge-aware loss function for ringing artifacts suppression.Comparative experiments show that EASRN effectively solves the problem of deblurring pixels with overexposure.In summary,this paper proposes innovative algorithms in three aspects:noise reduction,overexposure repair,and deblurring,which effectively solve the problem of image quality degradation of visible night images,thereby improving the performance of the imaging system.