Zoran Pecenovic,
Image retrieval using Latent Semantic indexing.
Final year graduate thesis, AudioVisual Communications Lab, Ecole Polytechnique
Fédérale de Lausanne, Switzerland, June 1997.
Zoran Pecenovic,
Image retrieval using Latent Semantic indexing.
Final year graduate thesis, AudioVisual Communications Lab, Ecole Polytechnique
Fédérale de Lausanne, Switzerland, June 1997.
Image Retrieval is a domain of increasing and crucial importance in the new information based society. Large and distributed collections of scientific, artistic, technical and commercial images are becoming a common ground, thus requiring sophisticated and precise methods for users to perform similarity and semantic based queries. Much work has already been done in this direction, but the majority is either too specific or too general. Encouraging results have already been achieved with semantic-based methods on textual information and have given us a starting point for the investigation of the feasibility of their application to image data. Latent semantic indexing has been successfully used in very large document collections, and its theoretical background allows for interesting possibilities. Latent semantic indexing extracts the underlying semantic structure, alleviates the variability, and reduces noise in ``term'' usage. The difficult question is of course: what kind of ``terms'' are images composed of? The work presented herein is the result of our attempt to partially and subjectively answer this question. After making a choice of ``terms'' we have implemented a prototype Latent semantic indexing image retrieval system. The collection of images was fairly large (approximately 4'500 images) and contained photographs and paintings of various very different subjects. Much importance was further given to query construction and relevance feedback in the system. The results were quite encouraging, and have lead us to a set of bright new outlooks on future improvements and extensions of the prototype.