Large Scale Image Search Benchmark

This work performs a thorough benchmark of the two leading approaches for large scale image search: Bag of Words (BoW) vs Full Representation (FR). It includes methods such as: Inverted File, Min-Hash, Kd-Trees, Hierarchical K-Means, Locality Sensitive Hashing (LSH), among others.


Details here

Online Parameter Selection for Large Scale Image Search

This work explores using online learning for selecting the best parameters of Bag of Words systems when searching large scale image collections.


Details here

Automatic Discovery of Image Families

This work investigates the problem of how to automatically discover image families in unorganized image collections. Image families are groups of images with significant similarity. The problem has applications to content-based image search, automatic visualization and organization of image collections, or computer vision research datasets, among others.


Details here

Real time Lane Detection in Urban Streets

This project was part of Team Caltech, Caltech's entry in the DARPA Urban Challenge in November 2007. This project's main aim was to detect and localize lane lines in urban streets, which will help Alice, Team Caltech's autonomous vehicle, find its way in traffic.


Details here

Fast Face Detection

This project was done as a requirement for the Computer Vision class EE148 of Spring 2006. The project was mainly to provide an open source Matlab implementation of a realtime face detection system developed by François Fleuret.

Details here

Face Recognition using SIFT Features

This project was part of the requirements of the CNS/Bi/EE 186: Vision: From Computational Theory to Neuronal Mechanisms class for Winter 2006. It implemented a simple face recognition system in Matlab exploiting the power of SIFT features to discriminate between faces of different individuals.

Details here

Survey on Multi-Class Classification

Description

This project aimed at compiling a survey for the various techniques used for multi-class classification. This is an important problem, specially in computer vision, where we would like to recognize hundreds of different object categories. The survey report introduces the several techniques employed to solve this problem, with a discussion of their advantages and disadvantages.

References

Mohamed Aly, Survey on Multi-Class Classification Methods. [pdf]