Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
IMAGE GENERATION APPARATUS AND IMAGE GENERATION METHOD
Document Type and Number:
WIPO Patent Application WO/2020/105144
Kind Code:
A1
Abstract:
This image generation apparatus is configured to comprise: an image generation unit (2) that uses a convolutional neural network having a plurality of convolution filters to generate, from an input signal, an image corresponding to a real image which is a generation target image; an image identification unit (3) that identifies whether or not the image generated by the image generation unit (2) is a real image; a filter coefficient update unit (4) that updates the filter coefficients of the plurality of convolution filters by using the identification result of the image identification unit (3); and a filter deletion unit (6) that deletes one or more convolution filters among the plurality of convolution filters on the basis of the importance of the plurality of convolution filters, wherein the image generation unit (2) generates, from the input signal, the image corresponding to the real image by using a convolution neural network having the remaining convolution filters, which are not deleted by the filter deletion unit (6), among the plurality of convolution filters of which the filter coefficients are updated by the filter coefficient update unit (4).

Inventors:
MIYAZAWA KAZUYUKI (JP)
MORIYA YOSHIMI (JP)
SUGIMOTO KAZUO (JP)
Application Number:
PCT/JP2018/043000
Publication Date:
May 28, 2020
Filing Date:
November 21, 2018
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
MITSUBISHI ELECTRIC CORP (JP)
International Classes:
G06T1/40; G06T7/00
Foreign References:
JP2018520404A2018-07-26
Other References:
"An illustrated guide to deep learning, revised second edition", 10 November 2018, KODANSHA LTD., JP, ISBN: 978-4-06-513331-6, article YAMASHITA, TAKAYOSHI: "Passages, An illustrated guide to deep learning, revised second edition", pages: 81-83,169 - 192, XP009521324
Attorney, Agent or Firm:
TAZAWA, Hideaki et al. (JP)
Download PDF: