基于构图分析的古代壁画相关度评价方法

Composition analysis-based relevance ranking for ancient mural

由于目前的图像检索技术没有考虑壁画的构图学特征,缺乏对复杂语义的处理能力,难以满足古代壁画研究工作对检索全面性和准确性的要求.为提高古代壁画图像语义检索的质量,提出基于构图分析的相关度模型,通过引入基于绘画构图学的理论和分析方法,从壁画内容的布局、主题和语义三方面用量化方法描述检索语义与壁画内容的相关度,较好地解决了用户的真实检索意图与壁画内容间的"语义鸿沟"问题.该相关度评价模型可嵌入基于语义查询扩展的框架中,以提高Top N结果的准确率,同时维持了较高的查全率.敦煌壁画资料检索的实际应用表明:以反映前n个结果准确率的R-Precision为评测指标,基于构图分析的相关度评价方法可比未采用相关度评价的基线方法平均高出36%.

The present image retrieval technologies have difficulties in retrieving ancient murals,since they lack of the abilities to handle complex semantic and features of layout in painting.This work puts forward a new relevance ranking model based on composition analysis to improve ancient mural retrieval.By introducing the theory of composition on painting,the relevance ranking model measures the relevance of mural images from three aspects which are layout,topic and semantics,and reduces the semantic gap between the content of mural and the real intention of the user.The relevance ranking model was seamlessly integrated into a unified framework for semantic query expansion to improve the precision of Top N results while maintaining a high recall.Experimental results of the Dunhuang Murals show that compared with the baseline method,the R-Precision ratio of semantic mural retrieval based on this model can be increased by 36% on average.