| 摘要 |
PURPOSE:To easily extract an object by extracting the object from an image while using occlusion information provided from the correspondence relation of plural images provided by a photographing system. CONSTITUTION:Human images 51 and 52 are formed on the respective image sensors of the two pairs of photographic systems and the mutual information of correspondent scanning lines of respective image sensors is inputted to a correspondent extraction block 53. The correspondent extraction block 53 extracts the correspondent points in the images of respective image sensors, but generates a part (occlusion part) where the correspondent points are not uniquely generated. In the case of the object having roundness like the human image, occlusion is generated along the contour of the object and therefore, the coordinate of this occlusion part is stored in an occlusion coordinate store part 54. Next, the scanning lines are updated, the similar processing is repeated, the correspondent points of images on all the scanning lines are extracted and on the other hand, the area of occlusion is provided. Thus, the human image can be easily extracted by detecting this occlusion area. |
