Straylights Suppression Of High-Reflective Objects Based On Multiband Polarization

Due to the influence of target's glossy surface and complex illumination conditions,the captured images are easily distorted by the disturbance of straylights during the imaging process,such as specular highlight and spotlight.The existence of specular straylights,may cover surface details,severely hindering some computer vision tasks,such as object detection,tracking,matching and 3D reconstruction.Existing straylights-removal methods can achieve convincing results to certain extent,while relying on too many priors and assumptions,such as dark channel prior.Meanwhile,the results inevitably suffer from color distortion,texture loss and boundary discontinuity.Also,these methods fail to handle the high-reflective objects with low structure and texture.To resolve the above two problems,through enormous experimental statistics and observations,we explore that the straylights-reduced results present strong structural similarity and color consistency in local region.The specular straylights region present distinct polarization features in different wavebands.Thus,we launch the research on straylights suppression of high-reflective objects based on multiband polarization imaging.The major contributions are listed as follow:1.In Chapter 2,to resolve the texture loss and boundary discontinuity in straylights removal process,a specular straylights removal method using local structural similarity and chromaticity consistency is proposed.The physical principles of specular straylights and diffuse reflection are presented,the dichromatic reflection imaging model is analyzed.Based on experimental statistics,the local structure similarity and chromaticity consistency are explored.Then,the specular straylights removal method using local structure similarity and chromaticity consistency regularization is proposed.Finally,experimental comparisons demonstrate that our algorithm can acquire reliable straylights-suppressed results,while dramatically preserving the local details,such as color,boundary.2.In Chapter 3,a straylights suppression method of high-reflective objects based on multiband polarization imaging is proposed,to effectively resolve the straylights of low-textured objects.The principles of multiband polarization imaging techniques are introduced,the multiband polarization distortion imaging model is established.Then,the straylights detection algorithm based on joint spectropolarimetric signatures constraint is presented,the Max-Min spectral differencing scheme,and ergodic least-squares separation are proposed and combined to remove the specular straylights.Also,to achieve better visual effect,the spatial color consistency regularization is exploited to compensate the missing details.Finally,experimental results validate the proposed method has the effectiveness and superiority to obtain reliable straylights-suppressed results qualitatively and quantificationally,especially for the low-textured and structured high-reflective objects.3.In Chapter 4,the multiband polarization-based straylights suppression and analysis system is established.The experimental platform is set up,and the straylights suppression and analysis software is designed and realized.Meanwhile,the high-reflective objects database is established,including multiple natural targets and man-made objects,in which the material properties and physical features are described.Experimental validations show the effectiveness and robustness of the proposed straylights suppression method in stereo matching application,pointing out the significance and engineering potentials of our work.