CN109657684A - A kind of image, semantic analytic method based on Weakly supervised study - Google Patents
A kind of image, semantic analytic method based on Weakly supervised study Download PDFInfo
- Publication number
- CN109657684A CN109657684A CN201811577772.6A CN201811577772A CN109657684A CN 109657684 A CN109657684 A CN 109657684A CN 201811577772 A CN201811577772 A CN 201811577772A CN 109657684 A CN109657684 A CN 109657684A
- Authority
- CN
- China
- Prior art keywords
- image
- region
- semantic
- weakly supervised
- subregion
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Abstract
The invention discloses a kind of image, semantic analytic methods based on Weakly supervised study, include the following steps;Step 1: dividing the image into the 45-55 blocks being of moderate size, the provincial characteristics with judgement index is extracted;Step 2: using linear classifier as the cluster mode differentiated;Step 3: the image area characteristics after segmentation are clustered, the image set after cluster is divided into the subregion of several classifications;It obtains a result Step 4: gathering same category of subregion in one kind;Step 5: the image-region of output semantic congruence merges, the final parsing result of image is obtained.The present invention provides a kind of image, semantic analytic methods based on Weakly supervised study, it is clustered by the image region that dividing method obtains, the Weakly supervised learning model minimized the error is established using image level mark and the relationship of image region label, for each image region allocated semantics label, it can achieve precision height, accuracy is high.
Description
Technical field
Automatically analyze and understand that technical field, specially one kind are based on Weakly supervised study the present invention relates to multimedia content
Image, semantic analytic method.
Background technique
Image, semantic parsing be divide the image into the combined task of area marking, be a kind of higher level
Image understanding technology, it not only can give image add semantic label, the correspondence area in image can also be added tags to
Domain realizes that more fine-grained image, semantic understands;
As internet enters the Web2.0 epoch, more and more users mark network image using semantic label
Note, and shared on picture sharing website Flickr, Picasa, explosive growth is presented in these image datas, to figure
The index of picture and retrieval bring huge challenge, for this purpose, quickly and effectively automatic image annotation becomes the hot spot of current research
Problem, both image segmentation and area marking are inseparable and mutually promote, and accurate image segmentation can be region
Mark provides accurate visual signature and indicates, conversely, good area marking result can equally promote image segmentation, because having
The pixel of same semantic label just belongs to the same object;
Image, semantic parsing is a kind of image labeling technology of thin scale, it will not only point out in image " what has ",
It is also noted that " where ", i.e., semantic label is mapped in image corresponding region up, to realize more careful accurate
Effect is marked, current existing image, semantic analytic method largely all relies on the training data accurately marked, i.e., artificial mark
The training image of pixel scale is infused, but the network image content change multiterminal of big data era, semanteme disperse different, consuming
The manual mask method of manpower increasingly cannot meet the needs.
Summary of the invention
The purpose of the present invention is to provide a kind of image, semantic analytic methods based on Weakly supervised study, to solve above-mentioned back
The problem of being proposed in scape technology.
To achieve the above object, the invention provides the following technical scheme: a kind of image, semantic solution based on Weakly supervised study
Analysis method, includes the following steps;
Step 1: dividing the image into the 45-55 blocks being of moderate size, the provincial characteristics with judgement index is extracted;
Step 2: using linear classifier as the cluster mode differentiated, and classifier is constrained using norm;
Step 3: the image area characteristics after segmentation are clustered, the image set after cluster is divided into several classes
Other subregion;
It obtains a result Step 4: gathering same category of subregion in one kind;
S1, object module is established with the corresponding the constraint relationship of image region label by image level label;
S2, the cluster set allocated semantics label by image region;
S3, input have the image of semantic label, are clustered using SHDC method to same category of subregion;
Step 5: the image-region of output semantic congruence merges, the final parsing result of image is obtained.
Preferably, in step 1, the feature in region can be extracted from color, texture and shape as feature.
Preferably, in step 1, the feature in region can be extracted from the significant point in image as feature.
Preferably, in step 1, the vision similarity between subregion is calculated by the way of norm reconstruct, is used simultaneously
The optimization method of CCCP and the iteration renewal process of non-negative multiplier method optimize norm item.
Preferably, in step 1, each image is split using the dividing method of SLIC.
Preferably, judged whether according to the size of color histogram map distance to image district when image-region is merged
Domain A and image-region B are merged.
Compared with prior art, the beneficial effects of the present invention are: the present invention provides a kind of figures based on Weakly supervised study
As semantic analytic method;
1, it is clustered by the image region that dividing method obtains, utilizes image level mark and image region mark
The relationship of label establishes the Weakly supervised learning model minimized the error, is each image region allocated semantics label, can achieve
Precision is high, and accuracy is high, just reaches higher semantic description precision using less parameter;
2, the context relation between region is taken full advantage of, information loss is reduced, using SHDC method to same class
Other subregion is clustered, and the effect of cluster is more preferable, can be measured respectively to object type sample and background classes sample, to increase
It is strong to the judgement index between object type and background classes, the needs of practical application, strong applicability can be met well.
Specific embodiment
Below in conjunction with the embodiment of the present invention, technical solution in the embodiment of the present invention is clearly and completely retouched
It states, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based on the present invention
In embodiment, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
Embodiment 1:
The present invention provides a kind of technical solution: a kind of image, semantic analytic method based on Weakly supervised study, including following
Step;
Step 1: divide the image into 48 blocks being of moderate size, using the dividing method of SLIC come to each image into
Row segmentation, extracts the provincial characteristics with judgement index, and the vision similarity between subregion is calculated by the way of norm reconstruct,
To the sensitivity of noise less, and reflection spatial information that can be implicit, it is more suitable for classification task, while using CCCP's
Optimization method and the iteration renewal process of non-negative multiplier method optimize norm item;
The feature in region can be extracted from color, texture and shape as feature, and color characteristic is the image bottom, most
Intuitive most apparent physical features, textural characteristics are the repetition distributions of certain approximate shapes, and shape feature is the outer boundary of object,
Extracted from color, texture and shape as feature have the characteristics that calculate simply express it is intuitive;
Step 2: using linear classifier as the cluster mode differentiated, and classifier is constrained using norm;
Step 3: the image area characteristics after segmentation are clustered, the image set after cluster is divided into several classes
Other subregion;
It obtains a result Step 4: gathering same category of subregion in one kind;
S1, object module is established with the corresponding the constraint relationship of image region label by image level label;
S2, the cluster set allocated semantics label by image region;
S3, input have the image of semantic label, are clustered using SHDC method to same category of subregion, use
SHDC method realizes more preferably Clustering Effect instead of traditional spectral clustering;
Step 5: the image-region of output semantic congruence merges, the final parsing of image is obtained as a result, in image district
According to the size of color histogram map distance when domain merges, judge whether to merge image-region A and image-region B.
Embodiment 2:
The present invention provides a kind of technical solution: a kind of image, semantic analytic method based on Weakly supervised study, including following
Step;
Step 1: divide the image into 50 blocks being of moderate size, using the dividing method of SLIC come to each image into
Row segmentation, extracts the provincial characteristics with judgement index, and the vision similarity between subregion is calculated by the way of norm reconstruct,
To the sensitivity of noise less, and reflection spatial information that can be implicit, it is more suitable for classification task, while using CCCP's
Optimization method and the iteration renewal process of non-negative multiplier method optimize norm item;
The feature in region can be extracted from the significant point in image as feature, and image can flexibly be depicted
Local message and detail content
Step 2: using linear classifier as the cluster mode differentiated, and classifier is constrained using norm;
Step 3: the image area characteristics after segmentation are clustered, the image set after cluster is divided into several classes
Other subregion;
It obtains a result Step 4: gathering same category of subregion in one kind;
S1, object module is established with the corresponding the constraint relationship of image region label by image level label;
S2, the cluster set allocated semantics label by image region;
S3, input have the image of semantic label, are clustered using SHDC method to same category of subregion, use
SHDC method realizes more preferably Clustering Effect instead of traditional spectral clustering;
Step 5: the image-region of output semantic congruence merges, the final parsing of image is obtained as a result, in image district
According to the size of color histogram map distance when domain merges, judge whether to merge image-region A and image-region B.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (6)
1. a kind of image, semantic analytic method based on Weakly supervised study, which is characterized in that include the following steps;
Step 1: dividing the image into the 45-55 blocks being of moderate size, the provincial characteristics with judgement index is extracted;
Step 2: using linear classifier as the cluster mode differentiated, and classifier is constrained using norm;
Step 3: the image area characteristics after segmentation are clustered, the image set after cluster is divided into several classifications
Subregion;
It obtains a result Step 4: gathering same category of subregion in one kind;
S1, object module is established with the corresponding the constraint relationship of image region label by image level label;
S2, the cluster set allocated semantics label by image region;
S3, input have the image of semantic label, are clustered using SHDC method to same category of subregion;
Step 5: the image-region of output semantic congruence merges, the final parsing result of image is obtained.
2. a kind of image, semantic analytic method based on Weakly supervised study according to claim 1, it is characterised in that: step
In one, the feature in region can be extracted from color, texture and shape as feature.
3. a kind of image, semantic analytic method based on Weakly supervised study according to claim 1, it is characterised in that: step
In one, the feature in region can be extracted from the significant point in image as feature.
4. a kind of image, semantic analytic method based on Weakly supervised study according to claim 1, it is characterised in that: step
In one, calculate vision similarity between subregion by the way of norm reconstruct, at the same using the optimization method of CCCP with it is non-
The iteration renewal process of negative multiplier method optimizes norm item.
5. a kind of image, semantic analytic method based on Weakly supervised study according to claim 1, it is characterised in that: step
In one, each image is split using the dividing method of SLIC.
6. a kind of image, semantic analytic method based on Weakly supervised study according to claim 1, it is characterised in that: step
In five, when image-region is merged according to the size of color histogram map distance, judge whether to image-region A and image district
Domain B is merged.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811577772.6A CN109657684A (en) | 2018-12-20 | 2018-12-20 | A kind of image, semantic analytic method based on Weakly supervised study |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811577772.6A CN109657684A (en) | 2018-12-20 | 2018-12-20 | A kind of image, semantic analytic method based on Weakly supervised study |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109657684A true CN109657684A (en) | 2019-04-19 |
Family
ID=66116291
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811577772.6A Pending CN109657684A (en) | 2018-12-20 | 2018-12-20 | A kind of image, semantic analytic method based on Weakly supervised study |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109657684A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115439688A (en) * | 2022-09-01 | 2022-12-06 | 哈尔滨工业大学 | Weak supervision object detection method based on surrounding area perception and association |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030147558A1 (en) * | 2002-02-07 | 2003-08-07 | Loui Alexander C. | Method for image region classification using unsupervised and supervised learning |
CN103336969A (en) * | 2013-05-31 | 2013-10-02 | 中国科学院自动化研究所 | Image meaning parsing method based on soft glance learning |
-
2018
- 2018-12-20 CN CN201811577772.6A patent/CN109657684A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030147558A1 (en) * | 2002-02-07 | 2003-08-07 | Loui Alexander C. | Method for image region classification using unsupervised and supervised learning |
CN103336969A (en) * | 2013-05-31 | 2013-10-02 | 中国科学院自动化研究所 | Image meaning parsing method based on soft glance learning |
Non-Patent Citations (1)
Title |
---|
文笃石: "基于二次聚类弱监督学习的图像语义分割", 《国外电子测量技术》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115439688A (en) * | 2022-09-01 | 2022-12-06 | 哈尔滨工业大学 | Weak supervision object detection method based on surrounding area perception and association |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | Integrating bottom-up classification and top-down feedback for improving urban land-cover and functional-zone mapping | |
CN109002834B (en) | Fine-grained image classification method based on multi-modal representation | |
CN102254192B (en) | Method and system for semi-automatic marking of three-dimensional (3D) model based on fuzzy K-nearest neighbor | |
TW201331772A (en) | Image index generation method and apparatus | |
CN102902826B (en) | A kind of image method for quickly retrieving based on reference picture index | |
CN104376105A (en) | Feature fusing system and method for low-level visual features and text description information of images in social media | |
CN108427713A (en) | A kind of video summarization method and system for homemade video | |
CN110399895A (en) | The method and apparatus of image recognition | |
CN112650923A (en) | Public opinion processing method and device for news events, storage medium and computer equipment | |
CN104636755A (en) | Face beauty evaluation method based on deep learning | |
CN108897778A (en) | A kind of image labeling method based on multi-source big data analysis | |
CN108898166A (en) | A kind of image labeling method | |
CN110472652A (en) | A small amount of sample classification method based on semanteme guidance | |
CN110378911A (en) | Weakly supervised image, semantic dividing method based on candidate region and neighborhood classification device | |
Madan et al. | Synthetically trained icon proposals for parsing and summarizing infographics | |
CN102646198B (en) | Mode recognition method of mixed linear SVM (support vector machine) classifier with hierarchical structure | |
CN110738033B (en) | Report template generation method, device and storage medium | |
CN106844785A (en) | A kind of CBIR method based on conspicuousness segmentation | |
CN114443855A (en) | Knowledge graph cross-language alignment method based on graph representation learning | |
CN110008365A (en) | A kind of image processing method, device, equipment and readable storage medium storing program for executing | |
CN106202391A (en) | The automatic classification method of a kind of user's community and device | |
CN103336830A (en) | Image search method based on structure semantic histogram | |
CN116882414B (en) | Automatic comment generation method and related device based on large-scale language model | |
CN108229565A (en) | A kind of image understanding method based on cognition | |
CN109657684A (en) | A kind of image, semantic analytic method based on Weakly supervised study |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |