Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
LEARNING METHOD AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2018/168539
Kind Code:
A1
Abstract:
The present invention includes: an input process for inputting a first image and a second image that constitute a moving image and are adjacent in time to a neural network, the second image lagging a prescribed time behind the first image; a learning process for causing the neural network to learn, using the first image and the second image, how to output a transformation matrix, used for all pixels in the first image, for transforming the first image to the second image; and an output process for outputting, as the result of estimation of motion between the first image and the second image, a movement amount image that represents the amount of movement until after a prescribed time of each pixel in the first image generated by the transformation matrix.

Inventors:
ALLETTO STEFANO
RIGAZIO LUCA
TSUKIZAWA SOTARO (JP)
Application Number:
PCT/JP2018/008230
Publication Date:
September 20, 2018
Filing Date:
March 05, 2018
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
PANASONIC IP CORP AMERICA (US)
International Classes:
G06T7/254; G06T7/00
Foreign References:
JPH11283040A1999-10-15
JPH04241077A1992-08-28
Other References:
LONG, GUCAN ET AL.: "Learning Image Matching by Simply Watching Video", EUROPEAN CONFERENCE ON COMPUTER VISION ECCV2016, 2016, pages 434 - 450, XP047355274, DOI: doi:10.1007/978-3-319-46466-4_26
G. LONGL. KNEIPJM ALVAREZH. LIX. ZHANGQ. YU: "European Conference on Computer Vision", vol. 1, 3, 2016, SPRINGER, article "Learning image matching by simply watching video", pages: 434 - 450
M. JADERBERGK. SIMONYANA. ZISSERMAN ET AL.: "Spatial transformer networks", ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS, vol. 2, 2015, pages 2017 - 2025
Attorney, Agent or Firm:
TOKUDA Yoshiaki et al. (JP)
Download PDF: