Statement
General Settings for feature matching still an open problem. An interesting perspective is the relation between the manifold learning theory and graph embedding and how this will reflect on a feature matching problem. The Work in the manifold learning theory includes techniques for dimensionality reduction; most of these techniques approximate the manifold by a graph to be embedded to a lower dimensionality space. Same reasoning can be applied to the problem of feature matching. The feature points of an image can be represented as a graph and hence can be embedded to a lower dimensionality space.

Not only the feature matching problem can be solved using the graph embedding of feature points, but also the object recognition and detection can be enhanced by using the manifold learning techniques on the feature points level. Exploring the relations, limitations and applications to the computer vision community is the core interest of my research.

My current research includes

Feature Matching

· I have developed a new framework for feature matching between multiple point sets depending on the spatial arrangement in each set and the feature similarity across different sets. Accepted for publication in CVPR2010.

Object Recognition and Detection

• Using a novel framework for embedding feature points from different images into the same coordinate system. The embedding framework preserves the spatial arrangement in each image and the feature similarity across different images.
• Kernel functions between images are established using the coordinates computed.
• To deal with large number of features and large number of images an out-of-sample embedding is proposed in a way that suits the object recognition context.
• The framework is applied on object recognition and detection successfully. Accepted for publication in CVPR2010.

Decision Support for Smart Trauma-Resuscitation Room TRU-IT: Trauma Resuscitation Unit -- Information Technology
Project web page
My role in the Project

· I helped in designing the environment that enable us to use the Stereo vision for applications of object detection and tracking objects that participate in the room.

· I helped in developing tracker for faces by working in the face detection step including training classifiers using Haar features.