Image Retrieval Systems based on Color Similarity and Edge Detection

This project was conducted in 2008 for my bachelor's thesis in the Department of Electrical Engineering at Seoul National University (it was more like a capstone project rather than a thesis, since the focus of the project was mainly at implementing algorithms rather than analyzing the performance of algorithms using concrete measures, e.g., recall and precision). I implemented an image retrieval system prototype that takes an image as input, and outputs most similar images from the image database. The measures used for the retrieval were color similarities: Earth Mover's Distance and Histogram Intersection. I compared the performance of the two metrics qualitatively by looking at the output images. The target image database was the Corel image dataset. The algorithms and GUI were implemented using MFC (Visual C++ 6.0) and OpenCV (image processing library). This project showed that histogram intersection was better in finding top-6 similar images (potentially high precision) than using Earth Mover's Distance for the target dataset. 

I presented this work to thesis committee members, and I received A+ in the "Bachelor's thesis" class (a project-based class advised by a professor and a Ph.D. student with no physical lectures). 

Thesis Committee: Dr. Namik Cho and Dr. Jungwoo Lee (Dept. of EE, SNU)