Image-based Object Recognition System
- Technology Benefits
- Outperforms state-of-the-art alternatives
- Technology Application
- Algorithmic recognition of an image of an unknown object from an image database, and in particular object recognition problems in which the linear feature model is valid.
- Detailed Technology Description
- None
- Supplementary Information
- Patent Number: US8406525B2
Application Number: US2010865639A
Inventor: Ma, Yi | Yang, Allen Yang | Wright, John Norbert | Wagner, Andrew William
Priority Date: 31 Jan 2008
Priority Number: US8406525B2
Application Date: 29 Nov 2010
Publication Date: 26 Mar 2013
IPC Current: G06K000966
US Class: 382191 | 382159 | 704236
Assignee Applicant: The Regents of the University of California | The Board of Trustees of the University of Illinois,Urbana
Title: Recognition via high-dimensional data classification
Usefulness: Recognition via high-dimensional data classification
Novelty: Method for recognition of high-dimensional data e.g. image of human face in presence of occlusion, involves identifying class of target data through linear superposition of sampled training data files by C1 minimization method
- Industry
- ICT/Telecom
- Sub Category
- Software/Application
- Application No.
- 8406525
- Others
-
Related Technologies
Tech ID/UC Case
17835/2007-115-0
Related Cases
2007-115-0
- *Abstract
-
Human faces are arguably the most extensively studied object in computer-based recognition -- due in part to the many important applications in this field as well as the realization that challenges associated with face recognition are representative of challenges in generalized object recognition. A central issue in research of object recognition systems is the question of which features of an object are most important for recognition. The dominant approaches are based techniques such as Eigenfaces, Fisherfaces, LaplacianFaces, and variants. However, with so many proposed features, there is a lack of guidelines for practitioners, and to-date, face recognition methods cannot achieve satisfactory results compared to human performance.
To address these challenges, researchers at UC Berkeley have examined this topic from the new context of the role of feature selection in face recognition from the perspective of sparse representation. This approach has led to an image-based recognition system that outperforms state-of-the-art alternatives. In testing, this approach achieve 96.5% recognition rate using 120 Eigenfaces on the entire Extended Yale B database. Other unconventional features such as severely down-sampled images and completely random projections performed equally well. For example, the holistic features given by images down-sampled to just 12 x 10 pixels achieved 92.4% recognition rate on the same database.
- *IP Issue Date
- Mar 26, 2013
- *Principal Investigator
-
Name: Yi Ma
Department:
Name: John Wright
Department:
Name: Allen Yang
Department:
Name: Andrew Wagner
Department:
- Country/Region
- USA
