Single Outdoor Image Depth Map Generation Based on Scene Classification and Object Detection

Three dimensional television (3DTV) has attracted more and more attention in the area of TV broadcasting. However, the applications are constrained due to the content shortage. It is an economical way by converting monoscopic 2D video to 3D (2D-3D) so as to reuse the existed huge amount of 2D videos materials by using Depth-Image-Based-Rendering (DIBR). In this paper, an efficient framework for extracting depth information from the single image is proposed, which is based on scene classification and object detection. In the proposed scheme, by applying that real similar 3D scenes may have a similar depth map, we construct an image set including many kinds of images (their corresponding depth maps are given) with different scene structures first. The image set is classified into some categories manually. For a certain input image, k-Nearest Neighbor (KNN) algorithm is employed to judge that whether the input image corresponds to outdoor scene or not. Then, the initial depth map is obtained by fusing the depth maps of those images in the category which the input image belongs to, After that, we incorporate the image segmentation results to detect the sky region and the ground region by using color information. Finally, the depth map is obtained by refining the initial depth map using the sky and ground region. Experimental results demonstrate that the proposed scheme can generate smooth and reliable depth maps with satisfied performance.