融合社交语义环境的网络图像标注关键技术研究

包含丰富社交语义信息的网络图像检索是当前社会多媒体知识发现的重要课题之一，本课题拟通过构建社交语义网络本体，并将其与传统的图像语义分析方法相结合，得到更为准确有效的图像标注模型。主要研究内容包括：（1）基于区域的图像内容表示方法研究，基于视觉注意力转移提出新的显著性分析方法和拓展显著性区域的图像内容表示模型；（2）图像语义分割研究，利用概率知识模型，深入分析图像视觉词及语义概念之间的关联，结合区域位置关系分析研究准确的图像语义分割算法；（3）图像多标签标注模型研究，利用图像不同区域不同层次的语义概念，采用多模型融合的方法构建标注新模型；（4）构建社交语义网络本体，提出用于社交网络图像语义理解的社交媒体协同标注新模型和新方法。本课题研究涉及多媒体信息处理、图像理解、模式识别、社会计算等多个学科，研究结果对于当前大规模网络图像的复杂语义内容分析及检索既有重要的理论意义，又有广泛的应用前景。

For reducing the influence of semantic gap on image understanding in social networks, the models and algorithms fused social semantic network ontology and traditional image semantic analysis methods are proposed for complex image semantic content analysis and retrieval. The main contents of research include: (1) Propose a new concept of moving visual focus for saliency analysis, and study the image content representation model based on extended saliency region. (2) Use probabilistic graphical model for In-depth analysis of the relationship between the visual features and semantic concepts of image. By combining with the ontology of region position, a more accurate model of image semantic segmentation is achieved.(3) For different regions and different semantic concepts levels of images, multi-model based multi-label annotation algorithms are proposed for complex semantic concepts annotation in terms of their different characteristics. (4) Make full use of the context information provided by social networks to construct the social semantic network ontology, and propose new computational models and methods of social media collaborative annotation for network image semantic understanding. Our study involves mutimedia information processing, image understanding, pattern recognition and social computing, which has important theoretical significance, and widely applications in the field of multimedia information retrieval.