Building Personal Image Retrieval Database with a MPEG-7 Based Crowdsourcing Annotation Methodology

• Main Authors: Cheng, Po Yu, 鄭博育
• Other Authors: Kuo, Pei Jeng
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/71621865053241960417

碩士 === 國立政治大學 === 數位內容碩士學位學程 === 101 === With the growing popularity of personal digital photography and video recording devices, the number of photographs and videos each person takes increases tremendously every year. However, storing such large quantity of digital files can cause major problems in archive management and retrieval. The occurrences of natural disasters and man-made accidents also contribute to the increasing trend of archive loss in data and photograph storage devices. Although images uploaded on web albums for storing and sharing can be retrieved using annotation tags, many network personal albums systems were found to have the following disadvantages: (1) lack of context-awareness tags; (2) lack of crowd-annotate tags functions, and (3) lack of annotated tags-search functions. Thus, this paper proposes a search architecture based on MPEG-7 and an expanded XML schema of the MPEG-7 architecture to describe photographic information and to provide richer information annotations for photograph retrieval. This study uses images files obtained from smart phones to examine image retrieval through image-annotated tags. Four forms of annotation tags were examined: automatic annotations, semi-automatic annotations and user-generated annotations by smart phone sensors, as well as crowd-generated annotations on Facebook.com.