Research On Key Technology Of Multi-view HDR Imaging

Comparing with low dynamic range(LDR)images,high dynamic range(HDR)images having wider dynamic range,more details and better visual effect overcome the problem of LDR images loss of detail in dark and light areas.Due to the high price of HDR capturing equipment,most HDR images are currently obtained by synthesizing multiple exposure images.However,there exist still many problems in the processes of synthesis.Multi-view high dynamic range image synthesis technology is studied in this thesis,which is focusing on image registration,virtual exposure image rendering and stereo ghost-free HDR image generation in the process of synthesis.The following three parts of the study:(1)The over-exposed images at highlight and the under-exposed images at dark lose details easily.It is difficult to obtain correct registration results,when over-exposure and under-exposure images with large differences in brightness are registered.To solve this problem,an image registration algorithm based on transform domain is proposed.Firstly,transforming images into the transform domain,and then registering by scale-invariant feature transform(SIFT).The influence of brightness is reduced in transform domain and more image features can be extracted,and finally better registration results can be obtained.The experimental results indicate that the registrations of the two images with large exposure differences in the transform domain are able to obtain better registration results.(2)Only one exposure image can be obtained per viewpoint when multi-view multi-exposure images taken at the same time,while the composite multi-view HDR image requires each viewpoint to have a set of multi-exposure images.For this problem,a virtual exposure image rendering algorithm based on multi-view multi-exposure image is proposed.Firstly,the camera response function(CRF)curve of the input image is estimated to unify the brightness between the input images,and then the disparity map between viewpoints is obtained by optical flow method,and rendered different exposure images to the same viewpoint.Then,in order to preserve more details and structure,it is necessary to fill the holes and repair the edges of the virtual exposure image,so the edge difference mask is introduced to correct and smooth the image edge information.Finally,we get the virtual exposure images.Experimental results show that the structure similarity between the rendered exposure image and the reference exposure view image is over 0.99.(3)For the problem that the ghost phenomenon generated when synthesizing single-view HDR images affects the quality of the final synthesized stereo HDR images,two methods of stereo HDR image synthesis based on ghost removal are proposed.While each viewpoint has a set of multi-exposure images,the CRF curve of the images is estimated,and then the corresponding response images of each viewpoint are obtained by CRF curve.The content of each response image is the same and the exposure is different.Then,registering multi-exposure images with the response images and synthesizing ghost-free HDR images,and ultimately obtaining stereo ghost-free HDR images.Secondly,if multiple viewpoints are shot at the same time,there is only one exposure image per viewpoint.Due to the difference of viewpoint positions,ghosts appear in the HDR images.By counting the brightness histogram of exposed images,the multi-exposure images of each viewpoint can be obtained through exposure transformation,and then the HDR image can be obtained.The results show that the algorithm can effectively remove the ghost phenomenon and realize the synthesis of stereo ghost-free HDR images.