CN105931253B - A kind of image partition method being combined based on semi-supervised learning - Google Patents

A kind of image partition method being combined based on semi-supervised learning Download PDF

Info

Publication number
CN105931253B
CN105931253B CN201610321142.7A CN201610321142A CN105931253B CN 105931253 B CN105931253 B CN 105931253B CN 201610321142 A CN201610321142 A CN 201610321142A CN 105931253 B CN105931253 B CN 105931253B
Authority
CN
China
Prior art keywords
image
gained
carried
node
split
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201610321142.7A
Other languages
Chinese (zh)
Other versions
CN105931253A (en
Inventor
马君亮
肖冰
汪西莉
何聚厚
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shaanxi Normal University
Original Assignee
Shaanxi Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shaanxi Normal University filed Critical Shaanxi Normal University
Priority to CN201610321142.7A priority Critical patent/CN105931253B/en
Publication of CN105931253A publication Critical patent/CN105931253A/en
Application granted granted Critical
Publication of CN105931253B publication Critical patent/CN105931253B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/001Image restoration
    • G06T5/003Deblurring; Sharpening
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20036Morphological image processing
    • G06T2207/20041Distance transform
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20112Image segmentation details
    • G06T2207/20152Watershed segmentation

Abstract

The invention discloses a kind of image partition methods being combined based on semi-supervised learning, include the following steps:Obtain the gradient image of image to be split, and treat segmentation image compressed, sharpened successively, binary conversion treatment and range conversion processing, obtain image to be split apart from topographic map;The gray value of each connected region is more maximum in extraction gained range conversion figure or point set, as foreground label;Watershed transform is carried out apart from topographic map to gained, using obtained watershed crestal line as context marker;The local minimum in the gradient image is shielded, the local minimum of the gradient image is marked according to the foreground of acquisition label and context marker, obtains revised gradient image;Then the acquisition of data of multiple angles, the foundation of prediction matrix, the structure of training pattern and the segmentation of image are carried out by semi-supervised learning method.The present invention can improve the precision of image segmentation.

Description

A kind of image partition method being combined based on semi-supervised learning
Technical field
The present invention relates to a kind of pattern dividing methods, and in particular to a kind of image segmentation being combined based on semi-supervised learning Method.
Background technology
Image segmentation, as an important branch in image procossing and computer vision field, is inhaled always with Objective extraction Draw the concern of numerous researchers.Meanwhile image segmentation and Objective extraction are in pattern-recognition, computer vision, artificial intelligence etc. Field also has a wide range of applications.Therefore, to the further investigation of image segmentation and Objective extraction not only facilitate image segmentation with The perfect solution of Objective extraction, and help to push the development in the fields such as pattern-recognition, computer vision, artificial intelligence.
Currently, image segmentation is mainly used to realize the classification of the data of unknown classification, in Analysis of Medical Treatment Data, credit card Credit is classified and there is great meaning in the fields such as image classification, once studying successfully and putting into application, will generate huge society Meeting and economic benefit.But the data (such as image in internet) in real world are no class label mostly, and sample This artificial calibration process is very time-consuming and laborious and expensive so that the Accurate classification of data acquires a certain degree of difficulty.Recently, it is based on similar The semi-supervised learning method of figure construction has been risen in the related fields such as data mining and pattern classification as powerful and popular Effective tool.Based on the data characteristics in real world, semi-supervised learning mainly by there is the classification of calibration sample, and has label Similitude between unlabeled exemplars discloses the classification of unlabeled exemplars
Traditional image partition method has:Mean shift process normalizes dividing method and K Mean Methods etc., generally deposits It is low in segmentation precision, target of the adhesion region without apparent boundary can not be divided, and be easy mutually to obscure with image background to cause to owe Segmentation.
Invention content
To solve the above problems, the present invention provides a kind of image partition method being combined based on semi-supervised learning, carry The high precision of image separation.
To achieve the above object, the technical solution that the present invention takes is:
A kind of image partition method being combined based on semi-supervised learning, is included the following steps:
The noise level of S1, extraction image to be split, and adjust according to gained noise level the bit of the image to be split Rate and resolution ratio, and with the bit rate and resolution compression of the gained image to be split;
S2, gray-scale map is generated according to the pixel edge strength of image obtained by step S1, and based on the gray-scale map to institute The Edge contrast of image is obtained, and obtains the gradient image of the image of gained;
S3, binary conversion treatment and range conversion are carried out successively to the image for sharpening gained, obtains the distance of image to be split Transformation Graphs, and complete the conversion of range conversion figure, obtain image to be split apart from topographic map;
The gray value of each connected region is more maximum in S4, extraction gained range conversion figure or point set, as foreground Label;Watershed transform is carried out apart from topographic map to gained, using obtained watershed crestal line as context marker;
Local minimum in S5, the shielding gradient image, according to gained foreground label and context marker label The local minimum of gradient image obtains revised gradient image.
S6, revised gradient image is divided into a certain number of region units, each image-region after segmentation is regarded For a node, three images of each region unit are established to the mapping of graphics, and are counted the syntople between vertex, calculated Connect the weight on the side on two vertex;
S7, all nodes are divided into mark node and do not mark node, wherein mark node occupies the minority;
S8, acquisition include the different visual angles sample data by various visual angles character representation for marking node and not marking node;
S9, the different visual angles sample data of gained is subjected to similarity-based learning, constructs similar neighborhoods figure, weight is calculated Coefficient matrix, and symmetrization, normalized are carried out to the weight coefficient matrix;
S10, it is marked according to the foreground, initializes a class label matrix;
S11, non-negative sparse is carried out based on the weight coefficient matrix after the class label matrix and symmetrization, normalized The iterative process that label is propagated, obtains prediction matrix;
S12, the similarity probabilities characterized according to gained prediction matrix, prediction do not mark node image different visual angles sample number According to accurate classification, obtain direct-push image classification as a result, training complete semisupervised classification modeling, generate training pattern;
S13, classification letter is carried out to the node image pattern to be sorted that do not mark in test set using gained training pattern The prediction of breath obtains the class label for not marking node image pattern to be sorted in test set, to realize to image data Cutting procedure.
Wherein, in the gray-scale map each pixel gray scale be video image in corresponding pixel points edge strength.
Wherein, the Edge contrast includes:
Expansive working and/or Gaussian Blur operation are carried out to the gray-scale map, obtain intermediate image A;
Etching operation is executed to the intermediate image A, obtains intermediate image A1;
Processing is sharpened to video image based on the intermediate image A1.
The invention has the advantages that:
Processing is sharpened to image by selectivity, is avoided doing over-sharpening;It is pre- to first pass through what range conversion obtained Foreground marks the context marker obtained with watershed transform to correct the gradient image of image to be split, then uses semi-supervised segmentation Method is split the revised gradient image, obtains image segmentation result, is carried out to gradient image so both having remained Watershed transform, can effective position target object edge, the advantages of being partitioned into target object integrity profile, and foreground mark can be passed through Note and context marker distinguish target of the adhesion region without apparent boundary so that the phenomenon that being not in less divided and over-segmentation, can To be suitable for the image segmentation field of object edge adhesion well, the precision of image segmentation is improved, promotion pattern is contributed to The development in the fields such as identification, computer vision, artificial intelligence.
Specific implementation mode
In order to make objects and advantages of the present invention be more clearly understood, the present invention is carried out with reference to embodiments further It is described in detail.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to limit this hair It is bright.
An embodiment of the present invention provides a kind of image partition methods being combined based on semi-supervised learning, including walk as follows Suddenly:
The noise level of S1, extraction image to be split, and adjust according to gained noise level the bit of the image to be split Rate and resolution ratio, and with the bit rate and resolution compression of the gained image to be split;
S2, gray-scale map is generated according to the pixel edge strength of image obtained by step S1, and based on the gray-scale map to institute The Edge contrast of image is obtained, and obtains the gradient image of the image of gained;
S3, binary conversion treatment and range conversion are carried out successively to the image for sharpening gained, obtains the distance of image to be split Transformation Graphs, and complete the conversion of range conversion figure, obtain image to be split apart from topographic map;
The gray value of each connected region is more maximum in S4, extraction gained range conversion figure or point set, as foreground Label;Watershed transform is carried out apart from topographic map to gained, using obtained watershed crestal line as context marker;
Local minimum in S5, the shielding gradient image, according to gained foreground label and context marker label The local minimum of gradient image obtains revised gradient image.
S6, revised gradient image is divided into a certain number of region units, each image-region after segmentation is regarded For a node, three images of each region unit are established to the mapping of graphics, and are counted the syntople between vertex, calculated Connect the weight on the side on two vertex;
S7, all nodes are divided into mark node and do not mark node, wherein mark node occupies the minority;
S8, acquisition include the different visual angles sample data by various visual angles character representation for marking node and not marking node;
S9, the different visual angles sample data of gained is subjected to similarity-based learning, constructs similar neighborhoods figure, weight is calculated Coefficient matrix, and symmetrization, normalized are carried out to the weight coefficient matrix;
S10, it is marked according to the foreground, initializes a class label matrix;
S11, non-negative sparse is carried out based on the weight coefficient matrix after the class label matrix and symmetrization, normalized The iterative process that label is propagated, obtains prediction matrix;
S12, the similarity probabilities characterized according to gained prediction matrix, prediction do not mark node image different visual angles sample number According to accurate classification, obtain direct-push image classification as a result, training complete semisupervised classification modeling, generate training pattern;
S13, classification letter is carried out to the node image pattern to be sorted that do not mark in test set using gained training pattern The prediction of breath obtains the class label for not marking node image pattern to be sorted in test set, to realize to image data Cutting procedure.
Wherein, in the gray-scale map each pixel gray scale be video image in corresponding pixel points edge strength.
Wherein, the Edge contrast includes:
Expansive working and/or Gaussian Blur operation are carried out to the gray-scale map, obtain intermediate image A;
Etching operation is executed to the intermediate image A, obtains intermediate image A1;
Processing is sharpened to video image based on the intermediate image A1.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art For member, without departing from the principle of the present invention, it can also make several improvements and retouch, these improvements and modifications are also answered It is considered as protection scope of the present invention.

Claims (3)

1. a kind of image partition method being combined based on semi-supervised learning, which is characterized in that include the following steps:
The noise level of S1, extraction image to be split, and according to gained noise level adjust the image to be split bit rate and Resolution ratio, and with the bit rate and resolution compression of the gained image to be split;
S2, gray-scale map is generated according to the pixel edge strength of image obtained by step S1, and based on the gray-scale map to gained figure The Edge contrast of picture, and obtain the gradient image of the image of gained;
S3, binary conversion treatment and range conversion are carried out successively to the image for sharpening gained, obtains the range conversion of image to be split Figure, and completes the conversion of range conversion figure, obtain image to be split apart from topographic map;
The gray value of each connected region is more maximum in S4, extraction gained range conversion figure or point set, as foreground label; Watershed transform is carried out apart from topographic map to gained, using obtained watershed crestal line as context marker;
Local minimum in S5, the shielding gradient image marks the gradient according to gained foreground label and context marker The local minimum of image obtains revised gradient image;
S6, revised gradient image is divided into a certain number of region units, each image-region after segmentation is considered as one A node establishes three images of each region unit to the mapping of graphics, and counts the syntople between vertex, calculates connection The weight on the side on two vertex;
S7, all nodes are divided into mark node and do not mark node, wherein mark node occupies the minority;
S8, acquisition include the different visual angles sample data by various visual angles character representation for marking node and not marking node;
S9, the different visual angles sample data of gained is subjected to similarity-based learning, constructs similar neighborhoods figure, weight coefficient is calculated Matrix, and symmetrization, normalized are carried out to the weight coefficient matrix;
S10, it is marked according to the foreground, initializes a class label matrix;
S11, non-negative sparse label is carried out based on the weight coefficient matrix after the class label matrix and symmetrization, normalized The iterative process of propagation, obtains prediction matrix;
S12, the similarity probabilities characterized according to gained prediction matrix, prediction do not mark node image different visual angles sample data Accurate classification obtains direct-push image classification as a result, semisupervised classification modeling, generation training pattern are completed in training;
S13, classification information is carried out to the node image pattern to be sorted that do not mark in test set using gained training pattern Prediction, obtains the class label for not marking node image pattern to be sorted in test set, is divided image data with realizing Cut process.
2. a kind of image partition method being combined based on semi-supervised learning according to claim 1, which is characterized in that institute The gray scale for stating each pixel in gray-scale map is the edge strength of corresponding pixel points in video image.
3. a kind of image partition method being combined based on semi-supervised learning according to claim 1, the Edge contrast Including:
Expansive working and/or Gaussian Blur operation are carried out to the gray-scale map, obtain intermediate image A;
Etching operation is executed to the intermediate image A, obtains intermediate image A1;
Processing is sharpened to video image based on the intermediate image A1.
CN201610321142.7A 2016-05-16 2016-05-16 A kind of image partition method being combined based on semi-supervised learning Active CN105931253B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610321142.7A CN105931253B (en) 2016-05-16 2016-05-16 A kind of image partition method being combined based on semi-supervised learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610321142.7A CN105931253B (en) 2016-05-16 2016-05-16 A kind of image partition method being combined based on semi-supervised learning

Publications (2)

Publication Number Publication Date
CN105931253A CN105931253A (en) 2016-09-07
CN105931253B true CN105931253B (en) 2018-11-06

Family

ID=56835930

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610321142.7A Active CN105931253B (en) 2016-05-16 2016-05-16 A kind of image partition method being combined based on semi-supervised learning

Country Status (1)

Country Link
CN (1) CN105931253B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107527349A (en) * 2017-06-27 2017-12-29 广州城建职业学院 Image partition method based on network dynamics evolutionary strategy
CN107967688B (en) * 2017-12-21 2021-06-15 联想(北京)有限公司 Method and system for segmenting object in image
CN108898618B (en) * 2018-06-06 2021-09-24 上海交通大学 Weak surveillance video object segmentation method and device
CN109460735B (en) * 2018-11-09 2021-02-02 中国科学院自动化研究所 Document binarization processing method, system and device based on graph semi-supervised learning
CN110517217B (en) * 2019-01-23 2020-04-28 莱芜职业技术学院 Computer integral data recognition device
CN111204046A (en) * 2019-04-09 2020-05-29 张柯 Saving type box body 3D printing system and method
CN109834941B (en) * 2019-04-09 2020-08-28 台州明创科技有限公司 Saving type box body 3D printing system
CN110119965A (en) * 2019-05-17 2019-08-13 浙江和也健康科技有限公司 A kind of mattress information collector, mattress after-sale service system and method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102013017A (en) * 2010-11-26 2011-04-13 华中科技大学 Method for roughly sorting high-resolution remote sensing image scene
CN102902978A (en) * 2012-08-31 2013-01-30 电子科技大学 Object-oriented high-resolution remote-sensing image classification method
CN104408731A (en) * 2014-12-09 2015-03-11 西安电子科技大学 Region graph and statistic similarity coding-based SAR (synthetic aperture radar) image segmentation method

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9129382B2 (en) * 2010-06-25 2015-09-08 Institute Of Automation, Chinese Academy Of Sciences Method and system for brain tumor segmentation in multi-parameter 3D MR images via robust statistic information propagation

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102013017A (en) * 2010-11-26 2011-04-13 华中科技大学 Method for roughly sorting high-resolution remote sensing image scene
CN102902978A (en) * 2012-08-31 2013-01-30 电子科技大学 Object-oriented high-resolution remote-sensing image classification method
CN104408731A (en) * 2014-12-09 2015-03-11 西安电子科技大学 Region graph and statistic similarity coding-based SAR (synthetic aperture radar) image segmentation method

Also Published As

Publication number Publication date
CN105931253A (en) 2016-09-07

Similar Documents

Publication Publication Date Title
CN105931253B (en) A kind of image partition method being combined based on semi-supervised learning
CN105608690B (en) A kind of image partition method being combined based on graph theory and semi-supervised learning
CN105389550B (en) It is a kind of based on sparse guide and the remote sensing target detection method that significantly drives
CN105095863B (en) The Human bodys' response method of semi-supervised dictionary learning based on similitude weights
CN103886342B (en) Hyperspectral image classification method based on spectrums and neighbourhood information dictionary learning
CN105956560A (en) Vehicle model identification method based on pooling multi-scale depth convolution characteristics
CN104239902B (en) Hyperspectral image classification method based on non local similitude and sparse coding
CN105469047A (en) Chinese detection method based on unsupervised learning and deep learning network and system thereof
CN103903013A (en) Optimization algorithm of unmarked flat object recognition
Xie et al. A deep-learning-based real-time detector for grape leaf diseases using improved convolutional neural networks
CN105335975B (en) Polarization SAR image segmentation method based on low-rank decomposition and statistics with histogram
CN109766835A (en) The SAR target identification method of confrontation network is generated based on multi-parameters optimization
CN102968796A (en) SAR (Synthetic Aperture Radar) image segmentation method based on sampling learning
CN104573699A (en) Trypetid identification method based on medium field intensity magnetic resonance dissection imaging
CN106846322A (en) Based on the SAR image segmentation method that curve wave filter and convolutional coding structure learn
CN108021869A (en) A kind of convolutional neural networks tracking of combination gaussian kernel function
CN104036296A (en) Method and device for representing and processing image
CN106548195A (en) A kind of object detection method based on modified model HOG ULBP feature operators
CN107944483B (en) Multispectral image classification method based on dual-channel DCGAN and feature fusion
CN103745233B (en) The hyperspectral image classification method migrated based on spatial information
CN105117740A (en) Font identification method and device
CN102436645B (en) Spectral clustering image segmentation method based on MOD dictionary learning sampling
CN102945553A (en) Remote sensing image partition method based on automatic difference clustering algorithm
Arbain et al. Offline handwritten digit recognition using triangle geometry properties
CN105005761A (en) Panchromatic high-resolution remote sensing image road detection method in combination with significance analysis

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant