Kazumasa Murai, Dr.

Kazumasa Murai is receiving Doctoral degree in computer science (PhD) from Chukyo University in 2006. I was a researcher with Fuji Xerox (1989-2000), with ATR Spoken Language Translation Laboratories (2000-2002), with FXPAL Japan, Fuji Xerox from 2002.

I am currently a senior researcher at FXPAL Japan, Fuji Xerox Co., Ltd. and a researcher at Institute for Advanced Studies in Artificial Intelligence, Chukyo University.

Born in July 28th, 1960, 45 year old male.

Research Topics

  • Multimodal speech prediction, detection and lip reading
  • Image analysis and object detection including face detection, object tracking, and 3D measurement
  • Application of neural network to image processing and color transformation.

Honors

Best research award from Imaging Society of Japan in 1995 (first author)

Best paper award from Imaging Society of Japan in 1996 (first author)

Odawara award from Japan Society of Mechanical Engineers in 2005 (coauthor)

Major fruits of my research

The most recent work is turn taking mediation system gAnticipatorh for the teleconferencing, which notifies the start and end of remote speech in advance, to reduce the speech collisions induced by the signal latency. One interesting point is that although the improvement is not so great, all of the examiners want to use the system because the system does not require any operation, yet helps the user to certain extent. This operation free user-interface is a natural outcome of my colleague and my standpoint as ordinary audiences of our own technologies.

Visual modality speech prediction is based on a fact that human form speech organs for articulation of the first phoneme prior to the speech. I have verified that people actually take advantage of this visual cue for smooth turn-taking. I have introduced very fist system that detects this precursor for the speech prediction.

Time backward multiple face tracking is also based on a very simple idea, which tracks faces (or any objects) in a video sequence against the timeline to recover more facial locations.

I also researched on accurate color transformation methods based on neural network. The basic idea is very simple, and I tried to formulate primitive phenomenon by simple mathematics , and that simple method drastically improved the color accuracy. I also invented an analog neural network for extremely fast color transformation. Although it is based on a relatively complex nonlinear circuit technology, the circuit itself is extremely simple.

Although my research results not always archived drastic improvement, I am always trying the idea to be used by reducing the user interface.

Publications and patents

Full paper (as a first author)

  • Kazumasa Murai, Hitoshi Ogatsu, Shinji Kita, gAn Accurate Color Transformation Algorithm based on Flexible GCR ?Accurate Printer Model by Neural Network-g J. Imaging Society of Japan, Vol.35, No.115, 1996
  • Kazumasa Murai, Satoshi Nakamura, gFace-to-Talk: Audio-Visual Speech Detection for Robust Speech Recognition in Noisy Environmenth, IEICE Trans. Info & Syst. Vol.E86-D, NO.3, 2003
  • Kazumasa Murai, Satosh Nakamura, gA Robust Bimodal Speech Section Detectionh, Journal of VLSI Processing Vol. 36, Kluwer, 2004
  • Kazumasa Murai, gVisual-Modality Speech and Conversation Modeling based on Facial Image Analysish, Doctoral Thesis, Chukyo University, 2006

Full paper (as a second or latter author)

  • Hitoshi Ogatsu, Kazumasa Murai, Shinji Kita, gAn Accurate Color Transformation Algorithm based on Flexible Gray Component Replacementh, Fuji Xerox Technical Report, Vol. 11, 1996.

International Conferences (as a first author)

  • Kazumasa Murai, Kenichi Kumatani, Satoshi Nakamura,h Speech Detection by Facial Image for Multimodal Speech Recognitionh, Proc. ICME 2001, 2001
  • Kazumasa Murai, Kenichi Kumatani, Satoshi Nakamura, gA Robust End Point Detection by Speaker's Facial Motionh, Proc. HSC2001 (International Workshop on Hands-Free Speech Communication), 2001
  • Kazumasa Murai, Satoshi Nakamura,h Real Time Face Detection for Multimodal Speech Recognitionh, Proc. ICME 2002, 2002
  • Kazumasa Murai, Don Kimber, Jon Foote Quiong Liu, John Doherty,h MEDIATED MEETING INTERACTION FOR TELECONFERENCINGh, Proc. ICME2005, 2005
  • Kazumasa Murai, gReal Time Multiple Face Detection and Tracking Method for a Meeting Captureh, Proc. QCAV2005, 2005

United States Patents (as a first author)

  • Kazumasa Murai, gData converting apparatus and coefficient determining apparatus for use therewithh, USP 5,568,414
  • Kazumasa Murai, Shinji Kita, Hitoshi Ogatsu, gMethod and apparatus for determining color transformation coefficientsh, USP5,978,107
  • Kazumasa Murai, Hitoshi Ogatsu, gColor image formation method and color image formation apparatush, USP 6,130,675
  • Kazumasa Murai, Hitoshi Ogatsu, gMethod and apparatus for processing a color imageh, USP 6,172,776
  • Kazumasa Murai, gData converting apparatus and coefficient determining apparatus for use therewithh, USP 6,304,863.

In addition, there are:

  • 24 Japanese patents as a first author,
  • 4 United States Patent as a second or latter author,
  • 13 Japanese patents as a second or latter author,
  • 6 Domestic (Japanese) conferences as a first author,
  • and many more publications as a second or latter author.