Large-Scale Multimodal Image Indexing and Retrieval with Lucene
Brown Bag Lecture by Dr. Matthew Simpson | 2/21/2012 11AM-12PM | 7th Floor Conference Room, Bldg 38A
Abstract: Images are sources of essential information within biomedical texts. However, given the rapid growth of biomedical literature, it is increasingly important to provide a means for quickly accessing the most relevant images for a given need. Whereas text-based retrieval systems utilize text describing an image, such as its caption, as a surrogate for its content, content-based and multimodal retrieval systems extract visual descriptors of an image’s content, such as its texture, and use these features to retrieve images that are visually similar to some given examples. Unfortunately, the use of content-based features becomes problematic at large scales when both the relevance of retrieved images and the time needed to obtain results are equally important.
This presentation describes a multimodal image indexing and retrieval approach suitable for use with large-scale image collections. The method operates by first clustering visual features extracted from a collection of images and then mapping the resulting clusters to artificial “words.” The method combines these words with other image-related text to create multimodal documents that it then indexes with a traditional text-based information retrieval system. Experimental results demonstrate that this approach allows for low latency retrieval and im¬proves upon the precision of existing content-based and multimodal methods.
Bio: Matthew Simpson rejoined the Lister Hill National Center for Biomedical Communications as a postdoctoral fellow in February 2011 after receiving his PhD in computer engineering from the University of Maryland, College park. He had previously obtained a BS in computer engineering from Clemson University in 2004 and worked at NLM while completing his graduate studies. Matthew is mentored by Dina Demner-Fushman, and his current research interests include biomedical information retrieval and natural language processing.