Image Retrieval With Embedded Region Relationships

Image retrieval based on content from digital libraries, multimedia databases, the Internet, and other sources has been an important problem addressed by several researchers. In this regard, one cannot overestimate the use of appropriate features such as color, texture, and shape. It has also become increasingly evident that the decomposition of images into regions is critical for useful results. In this paper we further study region-based image retrieval. We argue that a relationship between regions (such as a tiger amongst yellowish-green grass, or a plane against the blue sky with mountains in the background) is also important. Our local segmentation algorithm is used to detect regions a priori. Further, while searching for a match for an ‘object’ in the database, we allow for probabilistic ‘multiple matches,’ which are later pruned based on global consistent information. We provide a simple, fast algorithm implemented as an Internet thin client connecting to a web server. ...

• Type: Conference
• Year: 2003
• Where: SAC
• Authors: Sharat Chandran, Naga Kiran
• Keywords: Applied Computing, Image Retrieval, Local Segmentation Algorithm, Region-based Image Retrieval, SAC 2003