Title:
学習型アルゴリズムによるフォーカス補正処理方法
Document Type and Number:
Japanese Patent JP6727642
Kind Code:
B2
Abstract:
PROBLEM TO BE SOLVED: To propose a focus correction processing method by a learning type algorithm for processing a high-definition video image in real time.SOLUTION: A focus correction processing method is a step of estimating a parameter of the network in such a manner that the sum of the squared differences between a restored image which is a result of processing a smoothing input image by the network and a true image becomes the minimum so as to generate a smoothing input image assumed to be out of focus from a true image for using a convolutional neural network which is a learning type algorithm for a deblur restoration processing, and using such an input/output image as learning data. A learning parameter is regarded as the focus correction processing method by a learning type algorithm having a step of making the average ISNR of the deblur restoration processing in another estimation image different from the learning image become the maximum and a step of performing the focus correction processing using the convolutional neural network by the determined parameter that has finished learning.SELECTED DRAWING: Figure 1
Inventors:
Riki Matsunaga
Application Number:
JP2016090290A
Publication Date:
July 22, 2020
Filing Date:
April 28, 2016
Export Citation:
Assignee:
Toei Co., Ltd.
International Classes:
G06T1/40; G06N3/08; G06T5/00
Domestic Patent References:
| JP2008109640A |
Attorney, Agent or Firm:
Kazuhiro Kamata