Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
A picture division method, an image dividing device, a program, and a recording medium
Document Type and Number:
Japanese Patent JP6125099
Kind Code:
B2
Abstract:
An image segmentation method and device are provided. The image segmentation method comprises: a saliency model of an image is established (101); a foreground sample point and a background sample point of an image are acquired (102) according to the saliency model; a foreground and background classification model is established (103) according to the saliency model as well as the foreground sample point and the background sample point; the image is segmented (104) according to a predefined graph cut algorithm, and specifically the image is segmented according to the predefined graph cut algorithm by using the foreground and background classification model and edge information related to pixels. A foreground and background classification model is established by automatically determining a foreground sample point and a background sample point and by combining a saliency model, and image segmentation is achieved by using the foreground and background classification model; such a problem in the related technologies is solved that a user needs to manually and roughly select the foreground sample point and the background sample point and thus the segmentation efficiency is low for segmenting a large number of images; automatic selection of samples is realized and the accuracy of classification is improved.

Inventors:
One Lin
Ching Chiu Pin
Change
Application Number:
JP2016517158A
Publication Date:
May 10, 2017
Filing Date:
October 23, 2014
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
Xiaomi Incorporated
International Classes:
G06T7/40
Domestic Patent References:
JP2011150605A
Foreign References:
CN103208123A
Other References:
祗園 大輝 外3名,色顕著性に基づくインタラクティブ画像セグメンテーション,電気学会論文誌C Vol.133 No.6,日本,一般社団法人電気学会,2013年 6月 1日,第133巻 第6号,pp.1211-1217,ISSN 0385-4221
Attorney, Agent or Firm:
Ken Ieiri