MOTION BLUR REMOVING DEVICE, PROGRAM AND LEARNING METHOD

Title:

MOTION BLUR REMOVING DEVICE, PROGRAM AND LEARNING METHOD

Document Type and Number:

Japanese Patent JP2021111140

Kind Code:

A

Abstract:

To provide a motion blur removing device capable of effectively removing a motion blur.SOLUTION: A motion blur removing device 100 reads continuous frames on a video as an input tensor and obtains an output tensor in which a motion blur is removed with respect to one frame of them. The motion blur removing device is provided with a main net unit 10 configured to obtain the output tensor in which the motion blur is removed in consideration of time correlation and spatial correlation with respect to the input tensor, a sub-net unit 30 for extracting motion blur related information corresponding to one frame in the continuous frames by applying a convolution process after extracting a feature map including motion information from the input tensor, and an adjustment unit 20 for extracting an attention coefficient along the time axis of the input tensor from the feature map after extracting the feature map reflecting the motion blur information from the input tensor, in which the motion blur-related information and the attention coefficient are used for coupling and integration in the main net unit 10.SELECTED DRAWING: Figure 1

More Like This:

JPH11167628	PICTURE PROCESSOR
JP2021027565	IMAGE PROCESSING METHOD, IMAGING APPARATUS EMPLOYING THE SAME, IMAGE PROCESSING DEVICE, AND IMAGE PROCESSING PROGRAM
WO/1997/034253	FILTERING METHOD AND CORRESPONDING FILTERING SYSTEM

Inventors:

JO KENHO
TASAKA KAZUYUKI
CHOU YA

Application Number:

JP2020002903A

Publication Date:

August 02, 2021

Filing Date:

January 10, 2020

Export Citation:

Click for automatic bibliography generation Help

Assignee:

KDDI CORP

International Classes:

G06T5/00

Foreign References:

WO2017125980A1

2017-07-27

Other References:

WANG, XINTAO ET AL.: "EDVR: Video Restoration With Enhanced Deformable Convolutional Networks", 2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW), JPN6022050914, 16 June 2019 (2019-06-16), US, ISSN: 0004939044
柏原考爾: "畳み込みニューラルネットワークによる静脈画像の超解像処理", 電子情報通信学会技術研究報告, vol. 第116巻第196号, JPN6022050915, 18 August 2016 (2016-08-18), JP, pages 39 - 44, ISSN: 0004939045

Attorney, Agent or Firm:

Kiyotaka Sakamoto
Sanji Tanabe

Previous Patent: 情報処理装置、情報処理方法及び情報処理プログラム

Next Patent: 特徴ネットワーク抽出装置、コンピュータプログラム、特徴ネットワーク�...