Research On Ghost Removal Algorithm Based On High Dynamic Range Imaging Technology

With the rapid development of digital imaging technology,people's demand for digital image quality is gradually increasing.In recent years,because high dynamic range images contain higher contrast and more detailed information,they can better match the human eye's cognitive characteristics of brightness,saturation and contrast in real scenes,and have broad application prospects.It is widely used in digital imaging,biomedicine,telemetry and remote sensing,security monitoring and other fields.One method that is currently common is to obtain a high dynamic range image by multi-exposure image fusion technology.However,this method is only suitable for multi-exposure images taken in a static scene.Since most of the scenes in nature are dynamic,it is inevitable to include moving vehicles,crowds,etc.in different exposure images of multiple frames.Eventually,the high dynamic range image obtained by fusion produces a ghost effect,which greatly affects the image quality and visual effect.Therefore,it is crucial to study how to remove ghosts in high dynamic images in the field of image fusion.The thesis based on the research of high dynamic range imaging technology,ghost detection and removal technology,most of the existing advanced de-ghost algorithms need to perform motion estimation in the iterative optimization framework.This framework has high computational complexity and is not suitable for mobile devices.A ghost removal algorithm based on patch structure consistency detection is proposed to study the ghost detection and removal process.By detecting the similarity of the signal structure between the reference image patch and the corresponding position input multi-exposure sequence image patch,the structural inconsistency image patch is removed,and the purpose of removing ghost is achieved.At present,the drawback of the de-ghosting algorithm based on the reference image is that the pixel value of the ghost region of the final HDR image is determined only by the single-frame reference image.Once the overexposed/underexposed phenomenon occurs in some areas of the reference image,it will result in the lack of detail information in this part of the fused image.Based on this,the thesis first divides the reference image into two areas: normal exposure and underexposure/overexposure,and the two parts are processed in a targeted manner.For the subsequent patch fusion process,the thesis adopts a multi-exposure image fusion algorithm based on patch decomposition,which is to decompose the patch into three conceptually independent components: signal strength,signal structure and mean intensity,and each part is fused according to patch strength,exposure and structural consistency measures,then reconstruct the desired patch and add it to the final fused image.This approach not only does not require subsequent processing steps to improve image quality,but also collectively processes the RGB color channels of the image patchs to more naturally utilize color information.Finally,the proposed algorithm is compared with several advanced de-ghosting algorithms by experiments.The results show that the proposed algorithm can accurately detect ghost regions,and effectively remove ghosts while retaining the details of the exposed regions in the reference image,resulting in better visual effects and higher computational efficiency.