Research overview
With cheap cameras and ever-increasing digital storage capacity, we now have huge databases of images and video in personal collections and the web. Also, since cameras are inexpensive, they are increasingly used as a sensing mechanism for surveillance, robot navigation, and other applications. Analyzing the large amount of data requires efficient automatic techniques for extracting information from images. A primary task that lies at the heart of information extraction from visual data is image classification, which refers to classifying images or parts of them as belonging to certain categories. Accurate and reliable image classification can benefit us in diverse areas such as detecting suspicious activity in image sequences, medical image analysis, autonomous robotics, image search and content-based retrieval, automatic image annotation or tagging, etc. Most current approaches to image classification require large amounts of training data, usually provided through human labeling. However, providing training data needs tedious and time consuming human input. Especially for images and video, wherein scene conditions are varied, obtaining human training across the entire range of scene conditions is impractical. This training bottleneck is the motivation for development of robust algorithms that can perform good image classification with little human input.

In my Ph.D. research, I am working on novel algorithms for learning with image data. The focus is on active learning - instead of passively accepting training data, the basic idea in active learning is to actively select unlabeled examples for the human to label. The primary aim is to minimize human input in various kinds of image classification tasks that have traditionally required significant human supervision.

In prior work, I have studied how semi-supervised learning can be used for adaptive image classification when scene conditions change. In my Master's thesis, the developed techniques were applied for robust and efficient moving shadow detection in video sequences.

Publications