Component-Based Face Recognition
Face recognition technology is currently in high demand in security and surveillance markets. For Mitsubishi Electric to be competitive in these markets, world class face recognition software is needed. MERL has been working in earnest to develop a state-of-the-art face recognition system for over a year. Most recently we have improved the accuracy of our technology by using a component-based approach. Faces are divided into multiple regions (the eyes, nose and mouth) and independent recognizers are applied to each region and combined. The resulting combined component face recognition system is significantly more accurate than any single component recognizer.
Background & Objective: Our objective is to develop a state of the art face recognition system and to create some intellectual property for Melco in this area. In collaboration with Sentansoken, we have been comparing MERL's face recognition technology with OMRON's. This has lead to rapid improvement in MERL's system. The most important improvement is the development of a component-based face recognizer which has brought MERL's performance approximately to the same level as OMRON's. We expect further improvements to put us ahead of OMRON.
Technical Discussion: Our approach builds on our previous work at MERL on face detection which locates all faces in an image. We use the same machine learning framework developed for that problem and apply it to the problem of face recognition which determines the identity of a face. This requires extending the classifier to handle two input images at a time instead of just one. The classification problem is then to determine if the two faces are from the same person or not. The face recognition classifier is made more accurate by focusing on a single component (region) of the face at a time and then combining the independent results. The component-based recognizer also benefits from aligning each component to a particular facial feature. This effectively aligns faces more accurately without a global warping which would distort facial appearance.
| Technical Reports: | |
| Face Recognition Using Boosted Local Features | |
Technology Area: Computer Vision
Modification Date: July 7, 2005