▶ BK21 정보기술사업단-반도체공동연구소 공동세미나 안내 ◀
Biologically-Inspired Object Recognition
- 일 시 : 2011년 6월 27일 월요일 14:00-14:50
- 장 소 : 서울대학교 반도체공동연구소 설계연구관 3층 세미나
- 연 사 : 진용석 박사
Human visual system (HVS) is able to rapidly and effortlessly recognize diverse objects under different viewing conditions, such as changes in position, rotation and illumination. Neuroscientists have been researching on HVS and developing biologically-inspired object recognition models. HVS can be divided into two main phases, low-level (or early) vision and high-level vision. Low-level vision works as image preprocessing to extract various image features for various high-level vision tasks such as object recognition and salient region extraction.
Neuromorphic vision systems are hardware/software systems that implement biologically-inspired vision models that mimic the behavior of the visual cortex. In this talk, various biologically-inspired vision models will be introduced. Early visual processing consists of retina, lateral geniculate nucleus (LGN), and primary visual cortex (v1) parts, and extracts spatial and motion features. Attention-based on information maximization (AIM) is a saliency model to quantify visual saliency based on an information theoretic definition. Object recognition algorithms such as HMAX and HOP will also be introduced.
This talk also addresses implementation issues. Due to complexity and highly-dimensional design space, neuromorphic vision systems are generally slow. To implement a real-time neuromorphic vision system, hardware acceleration using a platform-based multi-FPGA vision system will be introduced. The neuromorphic vision systems are built on this platform which consists of configurable hardware vision operators, communication and peripheral modules, configuration tool, and performance analyzer.
Yongseok Jin is a postdoctoral researcher at the Pennsylvania State University. He received his Ph.D. in electrical engineering and computer science from Seoul National University in 2010. His research areas include multi-FPGA embedded system, video processing, and computer vision. He is leading Neovision2 project for developing a biologically-inspired vision system. He also contributed to the design of H.264/AVC video encoder and frame memory compression at Seoul National University.
담당교수 : 전기컴퓨터공학부 이혁재 (문의 : 김응섭, Tel. 880-1302 )