CN105512155A - Device and method for multi-layer semantic image retrieval - Google Patents
Device and method for multi-layer semantic image retrieval Download PDFInfo
- Publication number
- CN105512155A CN105512155A CN201410558869.8A CN201410558869A CN105512155A CN 105512155 A CN105512155 A CN 105512155A CN 201410558869 A CN201410558869 A CN 201410558869A CN 105512155 A CN105512155 A CN 105512155A
- Authority
- CN
- China
- Prior art keywords
- module
- image
- semantic
- target
- low
- 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
Abstract
The invention provides a device and method for multi-layer semantic image retrieval. The device comprises an information input module, an image retrieval module, a result output module and a display device, wherein the information input module, the image retrieval module and the result output module are connected in succession; the image retrieval module comprises a network communication module, a low-layer characteristic extraction module, a target semantic extraction module and a scene semantic extraction module, and the low-layer characteristic extraction module is connected to the network communication module; the target semantic extraction module is connected to the rear end of the low-layer characteristic extraction module; and the scene semantic extraction module is connected to the rear end of the target semantic extraction module, and retrieval results are displayed by the display device. The device and method provided by the invention have the beneficial effect that users' satisfaction with the image retrieval can be enhanced greatly.
Description
Technical field
The invention belongs to searching system field, especially relate to a kind of Multi-level semantic image indexing unit and method.
Background technology
Along with computing machine, the developing rapidly of multimedia and internet, internet there is the Digital image information resource of magnanimity.Picture search be in recent years growth rate the fastest classified search application, the picture search number of times of global a few large search engine is all doubled and redoubled.Therefore image retrieval technologies has become the focus of research both at home and abroad, and how efficiently tissue, the large-scale image data base of management and retrieval, also become the gordian technique in the major projects such as Future Information highway, digital library.Image retrieval is also a chief component of multimedia information retrieval technology, is one of theoretical foundation of Video Information Retrieval Techniques:, occupies very important status in information retrieval field.
Earlier picture retrieval, mainly through carrying out artificial textual annotation to image, utilizes text retrieval to realize searching characteristics of image.The mode that this employing manually marks keyword descriptor wastes time and energy, and subjectivity inevitably with a guy and uncertainty.The search engines such as current Baidu, Google automatically gather based on info web and label technology carries out text marking and retrieval to image.The image identification that this automatic marking gathers is very coarse, and accuracy is not high, sometimes or even inaccurate.Early 1990s, researcher proposes CBIR thought, achieves the leap of the retrievals of image vision content characteristic such as using color, texture, shape and region and the search modes of " to scheme to look for figure ".But people judge that the similarity of image is not only based upon on the similarity basis of these low-level features.CBIR only relate to the surface characteristics of image, the semantic meaning of the image that is beyond expression.
Enter 21 century, image retrieval launches around this focus of image, semantic, and its target is the understanding level making the ability of computer search image reach people.Ideally, user mainly carries out image retrieval according to the implication of image, instead of retrieves according to the low-level feature of image.
Image, semantic is broadly divided into Feature Semantics, Object Semanteme, Scene Semantics, behavior semanteme and emotional semantic etc., in order to be described the picture material of different levels.There is research that picture material is divided into three levels further: ground floor is primitive character layer, comprise the visual signature of Description Image, as color, texture, shape etc., reflection be the content that some of image have objective statistical characteristic, corresponding to the Feature Semantics of image; The second layer, for deriving attribute layer, relates to and being derived by Low Level Vision feature and the attribute that obtains, in order to the object (as " sun ", " basketball " etc.) described in recognition image, corresponding to the Object Semanteme of image; Third layer is abstract attribute layer, comprise the abstract attribute (as " sunrise ", " Basketball Match " etc.) that object and scene are carried out more high-rise reasoning and obtained, semantic and the emotional semantic of the Scene Semantics of correspondence image, behavior etc. generally the gap between ground floor and the second layer is called " semantic gap ", whether image retrieval really employs semanteme and is mainly reflected in the picture material whether obtaining the second layer.
Research in recent years based on the Images Classification of image, semantic, mark and retrieval mostly needs to utilize the strategy of machine learning to obtain semanteme, but does not occur a kind of gratifying system general efficiently yet.
Summary of the invention
The object of this invention is to provide a kind of Multi-level semantic image indexing unit and method, greatly can improve the satisfaction that user carries out image retrieval.
For achieving the above object, the technical solution used in the present invention is: a kind of Multi-level semantic image indexing unit and method, comprise MIM message input module, image retrieval module, result output module, display device, it is characterized in that: described MIM message input module, described image retrieval module, described result output module connects successively, described image retrieval module comprises network communication module, low-level feature abstract module, the semantic extraction module of target, and Scene Semantics extraction module, wherein, described low-level feature abstract module is connected with network communication module, and the low-level feature information of network image is extracted via network communication module, the semantic extraction module of described target is connected to the rear end of low-level feature abstract module, and in low-level feature information, extracts target semantic information according to predetermined probabilistic information, described Scene Semantics extraction module is connected to the rear end of the semantic extraction module of target, and in target semantic information, extract Scene Semantics information according to predetermined Scene Semantics template, again image corresponding for this Scene Semantics information is sent to the result output module that this Scene Semantics extraction module rear end connects, described display device display result for retrieval.
Preferably, described image retrieval module, based on the distance function of light stream histogram matrix, matches the video lens being similar to given video lens, retrieves matching image from video database.
Preferably, described display device is LED display.
The advantage that the present invention has and good effect are: by extracting target semantic information and the Scene Semantics information of network image, thus achieving the hommization image retrieval being different from prior art, substantially increasing the satisfaction of image retrieval.
Accompanying drawing explanation
Fig. 1 is structural representation of the present invention.
Embodiment
Below in conjunction with drawings and Examples, the present invention will be further described.
As shown in Figure 1, the invention provides a kind of Multi-level semantic image indexing unit and method, comprise MIM message input module, image retrieval module, result output module, display device, MIM message input module, image retrieval module, result output module connect successively, image retrieval module comprises network communication module, low-level feature abstract module, the semantic extraction module of target and Scene Semantics extraction module, wherein, low-level feature abstract module is connected with network communication module, and extracts the low-level feature information of network image via network communication module; The semantic extraction module of target is connected to the rear end of low-level feature abstract module, and in low-level feature information, extracts target semantic information according to predetermined probabilistic information; Scene Semantics extraction module is connected to the rear end of the semantic extraction module of target, and in target semantic information, extract Scene Semantics information according to predetermined Scene Semantics template, again image corresponding for this Scene Semantics information is sent to the result output module that this Scene Semantics extraction module rear end connects, display device display result for retrieval.
Preferably, image retrieval module, based on the distance function of light stream histogram matrix, matches the video lens being similar to given video lens, retrieves matching image from video database.
Preferably, described display device is LED display.
In the present invention, multiple keyword can be adopted to carry out image retrieval.Such as, for the network image preserving each regional aim semantic vector and integral image target semantic vector, when adopting scene keyword to retrieve, can directly according to the characters matching result in Scene Semantics layer, return image containing same scene semanteme in network image as result for retrieval.Or, when adopting target keyword to retrieve, according to the overall goals of network image semanteme return to should the image of target keyword as result for retrieval.Or, when the illustration adopting user manually to select is retrieved, select to there is the image of the target semantic vector matched as result for retrieval in network image according to the semantic proper vector of the target of this illustration.
The advantage that the present invention has and good effect are: by extracting target semantic information and the Scene Semantics information of network image, thus achieving the hommization image retrieval being different from prior art, substantially increasing the satisfaction of image retrieval.
Above embodiments of the invention have been described in detail, but described content being only preferred embodiment of the present invention, can not being considered to for limiting practical range of the present invention.All equalizations done according to the present patent application scope change and improve, and all should still belong within patent covering scope of the present invention.
Claims (3)
1. a Multi-level semantic image indexing unit and method, comprise MIM message input module, image retrieval module, result output module, display device, it is characterized in that: described MIM message input module, described image retrieval module, described result output module connect successively, described image retrieval module comprises network communication module, low-level feature abstract module, the semantic extraction module of target and Scene Semantics extraction module, wherein, described low-level feature abstract module is connected with network communication module, and extracts the low-level feature information of network image via network communication module; The semantic extraction module of described target is connected to the rear end of low-level feature abstract module, and in low-level feature information, extracts target semantic information according to predetermined probabilistic information; Described Scene Semantics extraction module is connected to the rear end of the semantic extraction module of target, and in target semantic information, extract Scene Semantics information according to predetermined Scene Semantics template, again image corresponding for this Scene Semantics information is sent to the result output module that this Scene Semantics extraction module rear end connects, described display device display result for retrieval.
2. a kind of Multi-level semantic image indexing unit according to claim 1 and method, it is characterized in that: described image retrieval module is based on the distance function of light stream histogram matrix, from video database, match the video lens being similar to given video lens, retrieve matching image.
3. a kind of Multi-level semantic image indexing unit according to claim 1 and method, is characterized in that: described display device is LED display.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410558869.8A CN105512155A (en) | 2014-10-18 | 2014-10-18 | Device and method for multi-layer semantic image retrieval |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410558869.8A CN105512155A (en) | 2014-10-18 | 2014-10-18 | Device and method for multi-layer semantic image retrieval |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105512155A true CN105512155A (en) | 2016-04-20 |
Family
ID=55720139
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410558869.8A Pending CN105512155A (en) | 2014-10-18 | 2014-10-18 | Device and method for multi-layer semantic image retrieval |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105512155A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491514A (en) * | 2017-08-07 | 2017-12-19 | 王庆军 | Image search method based on information security |
| CN109783749A (en) * | 2018-12-10 | 2019-05-21 | 深圳变设龙信息科技有限公司 | A kind of Material for design intelligent recommendation method, apparatus and terminal device |
| CN112651413A (en) * | 2019-10-10 | 2021-04-13 | 百度在线网络技术（北京）有限公司 | Integrated learning classification method, device, equipment and storage medium for vulgar graphs |
-
2014
- 2014-10-18 CN CN201410558869.8A patent/CN105512155A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491514A (en) * | 2017-08-07 | 2017-12-19 | 王庆军 | Image search method based on information security |
| CN109783749A (en) * | 2018-12-10 | 2019-05-21 | 深圳变设龙信息科技有限公司 | A kind of Material for design intelligent recommendation method, apparatus and terminal device |
| CN112651413A (en) * | 2019-10-10 | 2021-04-13 | 百度在线网络技术（北京）有限公司 | Integrated learning classification method, device, equipment and storage medium for vulgar graphs |
| CN112651413B (en) * | 2019-10-10 | 2023-10-17 | 百度在线网络技术（北京）有限公司 | Integrated learning classification method, device, equipment and storage medium for hypo-custom graph |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Snoek et al. | The challenge problem for automated detection of 101 semantic concepts in multimedia | |
| CN106560810B (en) | Searching using specific attributes found in images | |
| EP2402867B1 (en) | A computer-implemented method, a computer program product and a computer system for image processing | |
| US20100325138A1 (en) | System and method for performing video search on web | |
| CN101739428B (en) | Method for establishing index for multimedia | |
| CN103793697A (en) | Identity labeling method of face images and face identity recognition method of face images | |
| CN103384883B (en) | Top-K process is utilized to make semantic abundant | |
| CN102262670A (en) | Cross-media information retrieval system and method based on mobile visual equipment | |
| Dang-Nguyen et al. | Multimodal retrieval with diversification and relevance feedback for tourist attraction images | |
| CN102236714A (en) | Extensible markup language (XML)-based interactive application multimedia information retrieval method | |
| CN109446399A (en) | A kind of video display entity search method | |
| Kaneko et al. | Visual event mining from geo-tweet photos | |
| Fiallos et al. | Detecting topics and locations on Instagram photos | |
| CN113596601A (en) | Video picture positioning method, related device, equipment and storage medium | |
| Tsikrika et al. | Image annotation using clickthrough data | |
| CN105512155A (en) | Device and method for multi-layer semantic image retrieval | |
| CN103942274A (en) | Labeling system and method for biological medical treatment image on basis of LDA | |
| Min et al. | Multimodal spatio-temporal theme modeling for landmark analysis | |
| CN109241327A (en) | Image search method and device | |
| Tsai et al. | Qualitative evaluation of automatic assignment of keywords to images | |
| Boteanu et al. | Hierarchical clustering pseudo-relevance feedback for social image search result diversification | |
| Mylonas et al. | Context modelling for multimedia analysis | |
| Brenner et al. | Multimodal detection, retrieval and classification of social events in web photo collections | |
| CN201654783U (en) | Multi-level semantic image retrieval system | |
| Jones et al. | Automated annotation of landmark images using community contributed datasets and web resources |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160420 |
|
| WD01 | Invention patent application deemed withdrawn after publication |