Super-resolution Reconstruction Of Urban Streetscape Images Collected By The Vehicle

In order to ensure the high speed of images display and storage in real time,images captured by the popularization of driving recorder usually shows a low resolution.Generally speaking,the image sensor(CCD Charge-coupled device)can achieve better image quality,but on special occasions,due to imaging distance,imaging environment,the shape and the size of the sensor,the air disturbance,object motion and defocusing lens influence,making the images acquired by popularization of vehicle camera showing different types of degradation.such as noisy image,motion blurred image,lens distortion image,atmospheric turbulence blurred image and defocus blurred image.The degraded images can not effectively presented the details of image information.This will seriously affects the image information acquires when emergency situation is happening.In order to obtain sufficient image details(vehicle characteristics,Environmental characteristics)and improve the image quality,the super-resolution reconstruction technology has become a hot topic in the field of image processing.Super resolution reconstruction of image is of two main types: from the same scene of multiple degraded images to reconstruct a high resolution image;from a single image to reconstruct a high resolution image.Considering the nature of the image acquisition to the vehicle,this paper aims to solve the problem of super-resolution reconstruction of single vehicle collected images,which are images of environmental characteristics.Due to the influence of various factors,the most common degraded images of those five categories are noisy images and motion blurred images.So,this paper focuses on the research of the images of the urban streetscape images with the speckle noise and motion blur.First of all,this two kinds of degrade images need to be pretreatment to removal of the existing speckle noise or motion blur.Through the processing low resolution image of the vehicle can be acquired.And then the low resolution urban streetscape based on image collected by vehicle camera can be reconstructed to the high resolution image of self similarity in the images.For the reconstruction of vehicle image quality,the no-reference image quality assessment method is adopted to evaluate the quality of the images.Compared with the existing good super-resolution reconstruction algorithm this method of the image information entropy(entropy),image contrast(contrast),complexity(execution time),edge strength(ESL),blind image quality evaluation index(BIQI)have improved,proved effective this algorithm in vehicle image super-resolution reconstruction.The main work and innovation points are summarized as follows:(1)Analysis of different degraded images collected by vehicle.First it is important to eliminate image noise in the image preprocessing stage.We employ as feature the local weight function derived from steering kernel regression,and use principal component analysis(PCA)iterative training image feature of K-Means clustering,then a similar structure sub-dictionary was generated.With the method of kernel regression we can estimate image information and utilize total variation method to reconstructed image sharp edge;the last step is image fusion.(2)Analysis another kinds degraded images which have motion blur.According to the problem of estimating blur kernel not accurately,when the vehicular image has noise,a more accurate method of Variational Dirichlet distribution is proposed to estimate blur kernel,combined with improved Augmented Lagrangian achieve effective image restoration.This method uses the gradient projection method for solving optimization problems,extract precise orientation of the image edge blur after substitute posterior estimate eliminate image noise,reduce the additional constraint with Dirichlet distribution;hyper-Laplacian prior distribution modeling,carried out in conjunction ALM vehicular blind image restoration.(3)After image preprocessing of the urban streetscape images collected by vehicle camera.Images which are not containing noise and blur can be acquired.According to those images,the method we adopted is that the perspective transformation which based on self-examples of the images and high-frequency compensation were used to reconstruct the low resolution of the city street view images.The perspective transformation was added to the affine transformation through image patches matching,and for each matching image patches that high frequency compensation was used to recover the lost high frequency information when image pyramid was constructed.Through searching image pyramid of non-local multi-scale methods to get the patches which were used to synthetic images of high resolution.The experimental results show that: the super-resolution effect of the urban streetscape images collected by the vehicle is better than that of the current comparison algorithm from the subjective vision and the objective evaluation assessment.In the subjective visual effect,almost reconstruct images are no vignetting sharpening effect,image edge are clear;in the objective evaluation assessment,the information entropy of the image and contrast generally increase 0.5-2,edge strength and blind image reconstruction indicators generally increases 2-3,further verify the effectiveness of the proposed algorithm.