Semi-Automatic Annotation Of Personal Contents By Device Collaboration

Modern technologies allow consumers to create more and more personal digital contents. The size and type of personal digital contents are rapidly growing everyday. However, users still lack easy, intuitive and efficient methods to annotate and categorize their personal contents in a simple way. There are various reasons for this, such as lack of efficient content analysis methods, and the semantic gap between machine analysis and human interception. While one important reason that usually be ignored is that a single device typically has limited computational power and system resources, thus can only provide limited content analysis capabilities.To solve this problem we propose a method for interconnected devices to perform content annotation in a collaborative way. We design and implement an agent-like distributed content annotation protocol which enables one device to get help from other devices. These devices are allocated as a storage network by ad-hoc network links. This novel approach attempts to overcome the drawbacks of unbalanced computational power between devices with different configurations. It also enables the sharing of device knowledge in the form of training model sets for content analysis. In this way we can integrate all the system resources within the same network to perform a specific content analysis task, thus solve the unbalancing problem between different devices.We also study and implement a mechanism for user involvement in case proper analysis results cannot be achieved by automatic content analysis. By including user's feedback in the whole loop of content analysis, we would like to finally achieve a semi-automatic personal digital content annotation system which will provide a much better automatic content analysis performance and user experience.