Thomas de Buman

MSc Semester Project


JPSearch: Search engine for image databases


Thomas de Buman


Prof. Touradj Ebrahimi
Assistant: Dr. Frédéric Dufaux


2009, summer semester




Search engines are very important in order to be able to find relevant information in huge databases. Up to now, most developments have been on search of text documents (e.g. Google to search web pages). However, there is a growing interest in technologies for searching multimedia documents. To answer this demand, JPEG has started a new effort in order to standardize a norm, JPSearch, for the indexing and retrieval of still images.

More specifically, the goal of this project is to study Scale-Invariant Feature Transform (SIFT) and Speeded Up Robust Features (SURF) and to compare their performance for the search and retrieval of still images.

Precisely, the following tasks have to be preformed:

  • Experiment with SIFT features
  • Experiment with SURF features
  • Develop a search and retrieval demo system, taking a query image as input, and returning similar images from a database ranked by similarity
  • Compare the performance of SIFT and SURF for this task