融合上下文信息的场景结构恢复

提出了一种融合场景上下文信息的两级分类算法,从单幅图像中恢复场景结构。室外场景的结构化特征使其3维结构可以粗略地分为3类:"地面","天空"以及"竖直物体"。首先,把图像分割成具有灰度和颜色一致性的区域;其次确定特征显著区域("确定区域")的结构,将特征不明显的区域标记为"未知区域";然后根据"未知区域"与"确定区域"的相似性及"确定区域"场景结构对"未知区域"的可能结构进行投票,将投票最多的结构类型赋予"未知区域";最后介绍场景结构恢复在构造场景3维模型方面的应用。实验结果表明,由于利用了场景结构的上下文信息,该算法场景结构恢复的正确率为92.3%,优于现有算法88.1%的恢复正确率。

Structure recovery algorithm using contextual information

A two-level algorithm that is integrating contextual information to recover the structure of a single image is presented. Due to the structural features of outdoor scenes, we can classify the structure of a scene into three categories: sky, ground, and vertical objects. First, we over-segment the image into homogeneous regions. Then, we recognize the regions with significant features as "definite regions", and the regions we can not classify as "undetermined regions". Next, every nearby definite region with similar features as the undetermined region will vote for an undetermined region. The class with the most votes is assigned to that undetermined region. Finally, we construct a 3D model of the scene. Experiments show that due to the exploitation of the contextual information, almost 92.3% of the pixels can be recovered successfully, which is better than the performance of the existing method, whose result is 88.1%.