Research On Blind Restoration And Quality Assessment For Visible Sequence Images

In the measurement and control system of shooting range,the measurement data of the photoelectric measuring equipment is obtained through the analysis of optical image information.Therefore,the accuracy of measurement information is mainly determined by the quality of optical image.In the measurement and control experiment of shooting range,high-clarity and high-resolution images play an important role in the accurate information interpretation of target detection,tracking and recognition.In practical,the clarity of image is affected by a number of degradation factors.For optical-measuring devices,there are many factors leading to the degradation of image's quality,such as,the change of the optical measuring lens' s aperture and the optical system's transfer function under different application environment;the tracking jitter of optical measurement devices caused by the angle measurement error and the tracking error and so on;the sensor noise brought into the image;the loss of information during the image transmission process.The target light entering the optical-imaging system will be affected by the atmosphere,and this will result in a decline of image quality.For example,the changes of atmospheric temperature,water vapor,dust,aerosols,turbulence and other factors.Finally,the combination of these complex-degradation factors will result in blurred images and poor imaging quality,which seriously affects the acquisition of shooting range's high-accuracy measurement information.Therefore,it is urgent to improve the image quality and satisfy the acquisition of high-accuracy information.At present,obtaining high-quality images can be achieved in two ways.The first way is to change the hardware of existing equipment,add adaptive optics system,and improve the image quality by increasing the device's performance.But this solution is high-cost and difficult to achieve,it can only improve the performances of the devices,and can't change the influences from the external environment.The second way is to use image restoration technology to eliminate the noise and blurring,and estimate the high-quality images so that the image has a certain improvement in vision performances.This method is a low-cost and effective way to improve image quality.In this dissertation,aiming at the causes of degradation and the imaging characteristics of photoelectric devices,we study the image restoration technology which contains many kinds of noise and atmospheric degradation factors,and research the regularized image restoration technology.We also discuss how to restore more image details and how to remove image noise,so as to improve the image quality.In addition,the quantitative evaluation of the restored image is an important means to verify the effectiveness of the restoration algorithm.Therefore,we have also studied the quality evaluation of no-reference image.The main problems to be solved are as follows: 1)the restoration of atmospheric-degraded images with the influences from many kinds of noise;2)the influence of different image prior to the result of sequence images restoration;3)the application of correlation between sequence images in the restoration of sequence images;4)the application of the statistical feature in the no-reference image quality evaluation.The main research contents and innovations are as follows:1.An image degradation model based on the interaction of mixed noise and atmosphere is proposed,and a Bayesian estimation method is used to reconstruct the image.In this dissertation,the target images affected by atmospheric turbulence are studied.The causes of image degradation are analyzed and the image degradation model of mixed noise and atmosphere is established in the combination with the characteristics of target and image.In the solution,the adaptive total variation regularization term and the L1 blur kernel regularization term are used to restrain the clear image and the blur kernel respectively,and the high-resolution image solution is estimated by alternating the direction multiplier method.Experiments on multi-group images with different noise and atmospheric degradation are carried out to verify the effectiveness and applicability of the algorithm.2.A blind image restoration method based on non-rigid registration and graph Laplacian regularization term is proposed.The image registration and image deblurring are combined to restore the near-ground target image captured by the photoelectric imaging system.Image registration is used to eliminate geometric distortion between sequential images,and image deblurring is used to restore details and eliminate blur.In this dissertation,the graph theory is introduced.The graph Laplacian matrix is constructed by using the prior knowledge of image self-similarity,and its spectral characteristics are analyzed in detail.The normalized graph Laplacian matrix is used to construct a regular term with the characteristics of adaptive high-pass filter,and this regular term is set as the constraint of the image,so that it can play a better advantage in suppressing the noise and keeping the details of the image.Experiments with multiple sets of different target and degraded images of varying degrees are carried out to verify the effectiveness and applicability of the algorithm.3.A blind image restoration method based on adaptive spatial-temporal nonlocal total variation regularization terms is proposed.This dissertation further excavates the correlation between sequence images,generalizes the concept of space's nonlocal gradient to the time domain and elaborates the concept of spatial-temporal nonlocal gradient,so as to describe the change of nonlocal pixel values of sequence image in different spatial-temporal directions.In this dissertation,we propose an adaptive spatialtemporal nonlocal total variation regularization term to constrain high-resolution images with the combination of shooting range target's geometric characteristics,which can better recover the details of the image and estimate the optimal solution of the highresolution image.Experiments with multiple sets images of stationary and moving targets are carried out to verify the effectiveness of the algorithm.4.A no-reference image quality assessment method based on the characteristics of dual-tree complex wavelet statistics and support vector regression model is proposed.Dual-tree complex wavelet transform has many characteristics such as multidirectionality and multi-scale,it can better describe the geometric characteristics of shooting range targets.In this dissertation,the wavelet coefficients of each subband are obtained by the dual-tree complex wavelet transform,and the characteristic matrix(e.g.,amplitude,phase,mean value and variance of local entropy,phase entropy between scales)is extracted from the statistical feature.The characteristic matrix is regarded as the input to train and predict support vector regression model,and finally the quality of the image is evaluated.The correlation between the objective evaluation results and the subjective evaluation is tested to verify the effectiveness of the algorithm by the blurred and noisy image of the benchmark image dataset and the real atmosphere degraded image.