3D occluded object visualization by using integral imaging and semantic segmentation

Abstract

In this paper, we propose the improved occluded object visualization method by using integral imaging and semantic segmentation. Integral imaging is the passive 3D visualization technique that can generate 3D information through elemental images that have different perspectives of information on 3D objects. Moreover, it can be used to remove the occlusion in the 3D scene. The elemental image's various perspective information can be utilized to remove the occlusion in the 3D scene via the 3D image reconstruction process. However, the occlusion object pixels in the elemental image can degrade the image quality of the 3D image. Therefore, it is difficult to visualize the object without occlusion, clearly. To solve this problem, we propose the occluded object visualization method that can remove the occlusion and can visualize the target 3D object by using semantic segmentation. Semantic segmentation is the machine learning technique that can recognize the labeled object in the scene. Therefore, it can generate the specific labeled object mask image. Then, our proposed method can generate accurate 3D target object information. To prove our method, we carry out the simulation experiment and evaluate image quality with a correlation metric.