【摘要】 随着数字3D内容的不断丰富,对其进行有效展示和索引的需求变得十分迫切,而通过视点评分选择合适的观察视点对解决这一问题具有重要的作用.针对传统的视点评分方法计算时间冗长、结果很难符合人类观察习惯的问题,提出一种基于视觉感知信息量的快速视点评分方法.首先计算相对简单的平均曲率作为代表3D物体视觉特征的要素,随后结合信息熵理论对视点进行评分,使那些能看到尽可能多重要特征且所见特征分布较广的视点分数较高;之后,将这种评分策略应用于最优视点集计算中,利用信息理论对视点集获取信息进行量化,确保利用最少数目的视点有效地认知3D物体.实验结果表明,该方法得到的视点分数及最优视点集质量与目前最好的方法相当,但计算速度更快,无需人工交互.更多还原

【Abstract】 With the rapid growing of 3Ddigital contents,there is an urgent necessity for effective retrieval and displaying of the data.Viewpoint selection via viewpoint scoring is an important approach to the problem.However,traditional viewpoint scoring either requires complex computation,or fails to provide results that conform to human visual habits.To this,we propose a hybrid viewpoint scoring measure using visual perception and information entropy.In this paper,mean curvature is regarded as a key factor for visual perception of 3Dobjects.We score the viewpoints by blending visual perception and information entropy.Viewpoints observing more important features while the features are more widely distributed will be assigned higher scores.The scoring measure is then applied in the computation of best viewpoints,while the information acquired from the viewpoints is further quantized using information theory to ensure efficient perception of the 3D objects with minimum number of viewpoints.Experimental results show that the best viewpoints obtained by the proposed method better conforms to human visual habits.At the same time,the method has a higher speed and requires no user interaction.更多还原