Search International and National Patent Collections

1. (WO2018036293) IMAGE SEGMENTATION METHOD, APPARATUS, AND FULLY CONVOLUTIONAL NETWORK SYSTEM

Pub. No.:    WO/2018/036293    International Application No.:    PCT/CN2017/092614
Publication Date: Fri Mar 02 00:59:59 CET 2018 International Filing Date: Thu Jul 13 01:59:59 CEST 2017
IPC: G06T 7/00
Applicants: HANGZHOU HIKVISION DIGITAL TECHNOLOGY CO., LTD.
杭州海康威视数字技术股份有限公司
Inventors: SUN, Haiming
孙海鸣
XIE, Di
谢迪
PU, Shiliang
浦世亮
Title: IMAGE SEGMENTATION METHOD, APPARATUS, AND FULLY CONVOLUTIONAL NETWORK SYSTEM
Abstract:
Provided in an embodiment of the present application are an image segmentation method, apparatus and fully convolutional network system. Said method comprises: acquiring a target image to be processed; acquiring image feature data of the target image; inputting the image feature data into a target network which is obtained by pre-training and which is used for image segmentation, thereby obtaining an output result, wherein the target network is a fully convolutional network comprising a mixed context network structure, said mixed context network structure is used to integrate a plurality of reference features having pre-set scale ranges, which are extracted by the mixed context network structure itself, into target features, and the target features are features which match scale ranges of a target objects within a segmented image; the target network being formed by training with sample images of target objects having different scale ranges; obtaining an image segmentation result corresponding to the target image, according to the output result. By means of the present scheme, the efficiency of segmenting target objects of different scale ranges within an image can be increased while ensuring a large receptive field.