| 摘要: |
| 使用最大权匹配算法,结合统计平滑技术,提出图像区域特征生成概率估计方法,并进一步对训练集中标注词之间的语义相关性(correlation)进行分析与度量,给出一种基于生成模型的图像标注算法.算法使用所提出的基于最大权匹配的图像生成概率估计方法得到较好的起始点,进而设计启发式迭代函数对词与词的相关性加以利用,最终提高标注词与图像的语义相关性.在现实世界图像数据库上的实验结果验证了所提出标注方法的有效性. |
| 关键词: 图像标注 生成模型 连续特征估计 最大权匹配 语义相关性 |
| DOI： |
| 分类号: |
| 基金项目:Supported by the National Natural Science Foundation of China under Grant Nos.6040301, 860773077 (国家自然科学基金); the National Basic Research Program of China under Grant No.2005CB321905 (国家重点基础研究发展计划(973)) |
|
| Image Auto-Annotation via an Extended Generative Language Model |
|
WANG Mei,ZHOU Xiang-Dong,ZHANG Jun-Qi,XU Hong-Tao,SHI Bai-Le
|
| Abstract: |
| In this paper, based on the statistical smoothing strategy, a image region feature generative probability estimation method is proposed by exploiting maximum weight matching algorithm. By further analyzing and measuring the semantic correlations between words based on the training set, a novel image annotation algorithm for adopting the generative model is presented. The first annotation keyword is obtained by using the proposed image region feature generative probability estimation algorithm. Then, a heuristic iterate function is proposed to exploit the keyword semantic correlation. Finally, the semantic correlation between the annotation and the image can be improved by our annotation algorithm. The proposed annotation approach is tested on a real-world image database, and promising results are achieved. |
| Key words: image annotation generative model continuous feature estimation maximum weight matching semantic correlation |
