 « Other projects About the project "Colroy" is a project developed during the course Information Retrieval at the University of Massachusetts in the fall of 2006. Working on the project were Oskar Akerlund and Mattias Unger. The goal with this project was to create an image search engine based on color, using an image as a query. For testing purposes we needed a big collection of images which we got from Matton Images through the science project ImBrowse at Linkoping University. This collection includes over 750,000 images (25 Gb). With the great variety within the collection and a perfect candidate system for comparison, we set off implementing our ideas. Main ideas We divide each image into three grids of different accuracy. Within each sector of the grids we compute the average color in the CIELAB color space. The similarity function then depends on these colors. After indexing we are able to load the collection data into the memory which increases speed significantly, together with threads and compression. We built a stand-alone server that handles the index, receives queries from multiple clients and sends back sorted results. The client is a C# GUI where the user selects a query image or creates one by drawing. All parts of the projects are programmed by us from scratch. Results On a ordinary computer (see the report for specifications), our server searches through the collection in as fast as 0.2 seconds. The results (regarding relevant images) are also impressive compared to other image search engines, such as ImBrowse. Experiments and measurements are fully presented in our report below. Awards In MT Awards 2007, the jury’s honourable mention was given to Colroy by Mattias Unger and Oskar Akerlund with the explanation that “The project Colroy has, in a very impressive manner, tackled a complex problem within information retrieval. Colroy, through simple and smart methods, has created a powerful and very fast visual search portal whose results, in many ways, can be compared with larger and more established research projects.” Downloads Project report: report.pdf (3 Mb) |
© Oskar Akerlund 2003-2010