Research On High Dynamic Range Image Reconstruction Method

Humans mainly perceive and understand the unknown world by obtaining effective information,the visual system has always been an important way to obtain external information.With the development of digital information technology and the demand for human vision,imaging equipment has been greatly improved in terms of time resolution,spatial resolution,spectral resolution and dynamic response range.The imaging technology has gradually improved.In recent years,the imaging technology and its processing technology have played a vital role in various fields.However,due to the sensitivity of the device,the accuracy of the digital-to-analog converter and circuit noise,the images captured by conventional cameras can only record a limited dynamic range.Furthermore,due to the scene is unrepeatable and transient,the interested target cannot be captured again,and we can only process existing images.Therefore,reconstructing the high dynamic range image from low-quality images and improving the visual quality of scenes is a key issue in computer vision and has very important research value.This work is supported by the 973 Project,National High-Tech.(863)Project and State Key Program of National Natural Science Foundation et al..In this dissertation,we focus on the high dynamic range image reconstruction method in different scenarios,which includes improving image quality of static scenes,solving the de-ghosting problem when the scene has moving targets,image reconstruction in complex motion scenes and image reconstruction in the case of hand-held cameras.The main contribution of this thesis are summarized as follows:1.For the problem that the local details of imaging results are lost due to limited conditions,a high dynamic range image reconstruction method based on different exposure time is proposed.The proposed method achieves to restore high dynamic image in static scenes.Firstly,the imaging process of the camera is analyzed.From the camera imaging model and a large number of real data analysis,we propose that the brightness values of different exposure images vary linearly with the exposure time.Therefore,each of the low dynamic range images can generate more different exposed images.Secondly,in order to further discover the details and color information in the image,the gradient enhancement and white balance are introduced to process the images,which effectively enhances the details of the low-quality area and corrects the global color information.The experimental results show that the method of simulating different exposures is consistent with the actual situation,which can effectively improve the detail information of the image in the bright or dark area and enhance the visual effect of the image.Compared with other methods,the proposed method can handle general scenes.2.For the problem that the foreground motion causes ghosting in the results,a high dynamic range image reconstruction method based on the inter-frame attention mechanism is proposed to realize the high dynamic range image reconstruction without ghosting under dynamic scenes.This paper first introduces the attention mechanism to solve the ghosting problem which is caused by the difference between the non-reference frame and the reference frame content.By optimizing the feature of non-reference frame,the ghosting information is prevented from the source.In addition,during the reconstruction process,the dilation residual dense block is proposed.The dilation residual dense block makes full use of the features of different levels,retains more detailed information from the low dynamic range image,and increases the image receptive field to predict the details of the saturation region and motion region.The experimental results show that the proposed attention mechanism mitigates the impact of the motion region,and the local information is restored by using the dilation residual dense block.Under long-term exposure conditions,the proposed method is better than the existing algorithms and removes the ghosting artifacts.3.Ghosting and blurring often exit in reconstructed images due to large-scale motion and slight changes in background,a high dynamic range image reconstruction method based on low rank and optical flow is proposed to effectively apply to more complex scenes.In this paper,the low dynamic range image is mapped to the irradiance domain by the smoothness constraint of the CRF.Then two different algorithms are proposed to solve the problems in the foreground and the background respectively.We are the first to investigate the HDR image reconstruction under large-scale motion and the weak change.For the ghost region detection problem,an image de-ghosting method based on background low rank is proposed by using the priori information in the irradiance domain that background is low rank and the foreground is sparse.For the problem of background variation,an optical flow-based irradiance alignment method is proposed to align the background of the non-reference image to the background of the reference image.Finally,the results of the two are combined.The experimental results show that the proposed method not only ensures the high dynamic range image without ghosting,but also eliminates the background blur,which is more accurate than the traditional methods.4.Since hand-held camera captured image is often blurry and low dynamic range,a high dynamic range image reconstruction method based on multi-task network is proposed to recover HDR images from different exposed blurred images.Since camera shake can cause blurring during hand-held shooting,it is difficult for conventional methods to solve the coupled degraded images,even if sequential processing method.This paper innovatively adopts a two-branch network structure,and can simultaneously solve the problem of coupling of degradation factors.Each branch has both independent tasks and information transmission.The proposed network structure mainly includes two branches:image deblurring branch which recovers potential sharp image from input image;high dynamic range image reconstruction branch which recovers high dynamic range image from different blurred LDR inputs.In order to provide more comprehensive and clear features for high dynamic range imaging,this paper proposes different feature fusion methods,including multi-frame image feature fusion,multi-scale feature fusion and gate fusion,to guide the clear high dynamic range image reconstruction.