CN105404682A - Digital image content based book retrieval method - Google Patents
Digital image content based book retrieval method Download PDFInfo
- Publication number
- CN105404682A CN105404682A CN201510849994.9A CN201510849994A CN105404682A CN 105404682 A CN105404682 A CN 105404682A CN 201510849994 A CN201510849994 A CN 201510849994A CN 105404682 A CN105404682 A CN 105404682A
- Authority
- CN
- China
- Prior art keywords
- image
- book
- retrieval
- carrying
- carried out
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
Abstract
The invention discloses a digital image content based bood retrieval method. The method comprises the following steps: performing a resolution normalization operation on selected sample book images, carrying out multi-feature level image description, and then carrying out learning exercise on each type of books so as to obtain different training recognition models; correcting an obtained to-be-retrieved book image into a rectangle, then performing the resolution normalization operation on the image, performing multi-feature level image description after a corresponding image is obtained, inputting feature data into the training recognition model to perform prediction of initial results; and after the initial results are obtained, separately determining Gaussian distribution of two types of results with the maximum possibility and determining centers thereof, and determining the type of results with the highest probability as a final retrieval result. According to the retrieval method, bottom characteristics of the images are fully used, and by automatic learning and a model probability determination strategy, retrieval accuracy is improved and large-scale digital image based book retrieval work can be carried out.
Description
Technical field
The invention belongs to computer vision field, is an important application in field of image search, especially relates to a kind of book retrieval method based on digital image content.
Background technology
Fuzzy C-Means Cluster Algorithm is a kind of self-adaption cluster learning algorithm, and it can calculate characteristic of correspondence center automatically according to different classes of input training sample eigenwert, thus provides reliable priori for follow-up identification and testing.The advantage of this method is, characterizes comprehensively and work of can well classifying under the condition of sample size abundance at sample characteristics, and the tolerance power simultaneously for bad data is comparatively strong, can adapt to multiple different data environment.Therefore this algorithm is widely used in multiple image processing field such as data mining, Images Classification, image retrieval, image object identification, Video object tracking.The continuous increase of companion data amount in recent years, being applied in of this technology has promoted the development of computer vision technique to a certain extent.
Book retrieval based on image is an important application of image retrieval.It can retrieve corresponding book information fast according to image on the one hand, thus eliminates the link of the book information of input characters inquiry manually; Be enormously simplify the difficulty of information searching on the other hand by image retrieval, complete the search pattern of What You See Is What You Get.
Have some at present about the research of the book retrieval aspect based on image, also occurred some corresponding methods, wherein considerable method has all used characteristics of image and priori study.Divide from the book retrieval technical standpoint based on image, roughly can be divided into two large classes: the book image retrieval described based on characteristics of image and the book image based on information labeling are retrieved, and that pays close attention to herein focuses on the former.On the whole, book image based on information labeling is retrieved, forbidden due to markup information can be subject to, markup information can not meet the impacts such as all kinds of identification situations comprehensively effectively, the identification of these class methods is easily affected, and the environment adapted to also limit to relatively, does not have and effectively retrieves one to one, in addition because information labeling workload is difficult in maintenance greatly, institute receives application restriction in this way to a certain extent.Method accuracy rate comparatively speaking based on characteristics of image is higher, the multiple features of image describes the information such as edge, color, angle point effectively can analyzing books, the separatrix between different books can be found when training sample is enough large again by cluster analysis, in addition these class methods have stronger tolerance for the interference of extraneous data, can adapt to more identification scene.But the method based on training study needs the feature choosing comprehensive Description Image to obtain classifying face fully, in addition, the output for different cluster learning algorithm needs to find the acquisition that a kind of suitable result acquisition rule carries out net result.
With the development of graph image major field, the function of image retrieval can be more and more perfect, and requiring for book image retrieval can be more and more higher, but the environment of retrieval but becomes increasingly complex.Therefore make a kind of book retrieval recognizer that is reliable, robust to be extremely necessary.
Summary of the invention
The object of the invention is to the deficiency existed for prior art, propose a kind of book retrieval method based on digital image content.The method is based on the various features describing book image, incorporate the thought that clustering learning and probability distribution are established, substantially increase the efficiency of book retrieval, effectively can carry out the book image retrieval under different scene, have positive effect for image indexing system.
For achieving the above object, the invention discloses a kind of book retrieval method based on digital image content, the method comprises the following steps:
The sample image selected is carried out resolution normalization operation, and carries out the iamge description of multiple features aspect, then learning training is carried out to often kind of books, obtain different training model of cognition with this;
The book image to be retrieved obtained is corrected as rectangle, then resolution normalization is carried out to it, it is carried out to the iamge description of multiple features aspect after obtaining out corresponding image, characteristic is inputted the prediction carrying out initial results in described training model of cognition; Described predict the outcome in comprise the first two result of current books maximum possible to be retrieved;
After obtaining initial predicted result, judge two class results of maximum possible and the Gaussian distribution at its differentiation center respectively, that class that probable value is the highest, is defined as final result for retrieval.
Preferably, the iamge description of multiple features aspect comprises: describing the color moment feature of book image integral color distribution, describing one or more for describing in the Corner Feature of book image angle point of the edge histogram characteristic sum at entire image edge;
Preferably, sample image all carries out resolution normalization operation when carrying out book image feature extraction.
Preferably, sample image is when carrying out color moment feature extraction, after resolution normalization operation terminates, first carry out gamma filtering, then sample picture breakdown is become to have overlapping image block, wherein every two adjacent image block Duplication are 0.3 to 0.5 times of its area, then the color feature value of respective block in computed image, and finally statistics exports the colouring information of entire image.
Preferably, sample image, when carrying out color moment feature extraction, first carries out piecemeal process to sample image.
Preferably, sample image, when carrying out Corner Feature and extracting, carries out gray processing process to coloured image after resolution normalization operation terminates.
Preferably, sample image is when carrying out edge histogram feature extraction, after resolution normalization operation terminates, gaussian filtering is carried out to each color channel image in coloured image, picture breakdown is become to have overlapping image block, wherein every two adjacent image block Duplication are 0.3 to 0.5 times of its area, then the gradient distribution situation of statistical picture within the scope of 360 degree in each image block, finally statistics exports the marginal information description value of entire image.
Preferably, after obtaining each category feature of Description Image, with Fuzzy C-Means Cluster Algorithm, clustering learning is carried out to sample characteristics, will merge to obtain overall training study parameter to selected sample characteristics data before training study.
Preferably, the prediction steps of carrying out initial results comprises:
Obtain before initial results in calculating, carry out center traversal according to the breadth-first strategy of tree and calculate similarity, maximum two classes of last similarity are by as initial candidate result.
Preferably, training model of cognition is by the learning algorithm based on Clustering, obtains after carrying out learning training to often kind of books.
The present invention can the comprehensively color of Description Image, angle point and marginal information, and characteristic clustering technique can have stronger tolerance to the interference of extraneous data, and can adapt to multiple identification scene.Differentiate theoretical for accurate fusion results have employed gaussian probability after acquisition initial predicted result, effectively raise the accuracy rate of book retrieval.
Accompanying drawing explanation
A kind of book retrieval method flow schematic diagram based on digital image content that Fig. 1 provides for the embodiment of the present invention;
The color moment feature extracting method schematic process flow diagram that Fig. 2 provides for the embodiment of the present invention;
The edge histogram feature extracting method schematic process flow diagram that Fig. 3 provides for the embodiment of the present invention;
The Corner Feature extracting method schematic process flow diagram that Fig. 4 provides for the embodiment of the present invention.
Embodiment
By reference to the accompanying drawings, the basic thought of the embodiment of the present invention is the actual conditions for book image retrieval, and whole identification work is divided into three parts.Before carrying out book image retrieval, first rectangle rectification, normalization are carried out to book image and carry out series of features extraction and the learning and training based on Clustering again, to obtain the feature clustering center in book image storehouse; Then enter rectangle rectification, row normalization and feature extraction at cognitive phase again according to the book image to be retrieved obtained, gained feature is put into book image library model and obtains initial results; Finally merge to obtain last result for retrieval to initial results according to gaussian probability distribution theory again.Above method can adapt to multiple identification scene, and improves accuracy of identification in suitable degree.
A kind of book retrieval method flow schematic block diagram based on digital image content that Fig. 1 provides for the embodiment of the present invention.As shown in Figure 1, the method comprising the steps of 101-103:
In step 101, the sample image selected is carried out resolution normalization operation, and carry out the iamge description of multiple features aspect, then learning training is carried out to often kind of books, obtain different training model of cognition with this.
Particularly, in the learning training stage of book image retrieval, first rectangle rectification is carried out to corresponding sample image, resolution normalization operates, extract the edge histogram feature (extracting method process flow diagram as shown in Figure 2) of Description Image integral edge information again, the color moment feature (extracting method process flow diagram as shown in Figure 3) of Description Image colouring information and the Corner Feature (extracting method process flow diagram as shown in Figure 4) for Description Image angle point, then Category learning is carried out by the learning algorithm based on Clustering, be intended to for follow-up concrete identifying provides good differentiation model of cognition.
Preferably, sample image is when carrying out multiclass feature extraction, and resolution, according to the original scale of image, is normalized to suitable size; Sample picture is when carrying out color feature extracted, first gamma filtering will be carried out after resolution normalization terminates, then picture breakdown is become to have overlapping image block, wherein every two adjacent image block Duplication are 0.3 to 0.5 times of its area, with the impact that opposing is caused retrieval by the rotation translation of image, then the color feature value of respective block in computed image, finally statistics exports the colouring information of entire image; Sample image, when carrying out Edge Gradient Feature, will carry out gaussian filtering to each color channel image in coloured image after resolution normalization terminates, and the object done like this reduces the edge-description error brought by Gaussian noise.Equally picture breakdown is become to have overlapping image block in this link, wherein every two adjacent image block Duplication are 0.3 to 0.5 times of its area, then the gradient distribution situation of statistical picture within the scope of 360 degree in each image block, finally statistics exports the marginal information description value of entire image; Sample image, when carrying out Corner Feature and extracting, will carry out gray processing process to it to improve arithmetic speed after image normalization; When carrying out Corner Feature and extracting, after resolution normalization operation terminates, gray processing process is carried out to coloured image.
One as the embodiment of the present invention is improved, the embodiment of the present invention is after obtaining each category feature of Description Image, with Fuzzy C-Means Cluster Algorithm, clustering learning is carried out to sample characteristics, will merge to obtain overall training study parameter to selected sample characteristics data before training study.By the learning algorithm of fuzzy C-means clustering thought, learning training is carried out to often kind of books, obtain different training model of cognition.
In step 102, the book image to be retrieved obtained is corrected as rectangle, then resolution normalization operation is carried out to it, it is carried out to the iamge description of multiple features aspect after obtaining out corresponding image, characteristic is inputted the prediction carrying out initial results in described training model of cognition; Described predict the outcome in comprise the first two result of current books maximum possible to be retrieved.
Particularly, in the retrieval stage of book image retrieval, first, rectangle rectification is carried out by taking the irregular books obtained according to corresponding rectangle correction algorithm, then resolution normalization is carried out to it, extract corresponding color moment feature, edge histogram characteristic sum Corner Feature again, finally will enter the prediction of initial results in these characteristics input model of cognition; Comprise the highest front two class results of current book image to be detected in predicting the outcome, be used for reducing the identification range of corresponding book image.
Preferably, the embodiment of the present invention obtains before initial results in calculating, carries out center traversal according to the breadth-first strategy of tree and calculates similarity, and maximum two classes of last similarity are by as initial candidate result.
In step 103, after obtaining initial predicted result, judge two class results of maximum possible and the Gaussian distribution at its differentiation center respectively, that class that probable value is the highest, is defined as final result for retrieval.
Particularly, in the process that final result for retrieval differentiates, first according to the characteristic attribute of input amendment, merged by gaussian probability location mode and differentiate the probable value of current sample in all kinds of, finally obtain the accurate output predicted the outcome, wherein corresponding book information is maximum that of probability results predicted value.
Require book image comparatively clear in this course and area shared by books should be comparatively large, this is very important to effectively carrying out book retrieval.Being BFS (Breadth First Search) method based on center tree with what consider during feature decision that the operational performance of algorithm adopts.The probability output method of result should be used in last result differentiates.
Preferably, the embodiment of the present invention, when obtaining initial results, needs to carry out similarity measurement according to the method for BFS (Breadth First Search); After obtaining initial results under the condition of gaussian probability distribution, probability is carried out to it and differentiates to obtain final result.
The embodiment of the present invention, on the basis of image characteristics extraction, is combined togather image low-level image feature, Clustering and the probability forecasting method applied in image processing field widely effectively.The training study of Part I can obtain model of cognition according to selected books sample image.At cognitive phase, carry out merging, until obtain net result based on the result of Probabilistic according to the concrete initial predicted result obtained.
The present invention can under the prerequisite having ensured basic recognition function, and structure is simple, and complexity is low, and efficiency of algorithm is high, is adapted at applying in Books Retrieve System.
Obviously, under the prerequisite not departing from true spirit of the present invention and scope, the present invention described here can have many changes.Therefore, all changes that it will be apparent to those skilled in the art that, all should be included within scope that these claims contain.The present invention's scope required for protection is only limited by described claims.
Claims (10)
1., based on a book retrieval method for digital image content, it is characterized in that, comprise the following steps:
The sample image selected is carried out resolution normalization operation, and carries out the iamge description of multiple features aspect, then learning training is carried out to often kind of books, obtain different training model of cognition with this;
The book image to be retrieved obtained is corrected for rectangle, then resolution normalization operation is carried out to it, it is carried out to the iamge description of multiple features aspect after obtaining out corresponding image, characteristic is inputted the prediction carrying out initial results in described training model of cognition; Described predict the outcome in comprise the first two result of current books maximum possible to be retrieved;
After obtaining initial predicted result, judge two class results of maximum possible and the Gaussian distribution at its differentiation center respectively, that class that probable value is the highest, is defined as final result for retrieval.
2. method according to claim 1, it is characterized in that, the iamge description of described multiple features aspect comprises: describing the color moment feature of book image integral color distribution, describing one or more for describing in the Corner Feature of book image angle point of the edge histogram characteristic sum at entire image edge.
3. method according to claim 1, is characterized in that, described sample image all carries out resolution normalization operation when carrying out book image feature extraction.
4. method according to claim 2, it is characterized in that, described sample image is when carrying out color moment feature extraction, after resolution normalization operation terminates, first carry out gamma filtering, then become to have overlapping image block by sample picture breakdown, wherein every two adjacent image block Duplication are 0.3 to 0.5 times of its area, then the color feature value of respective block in computed image, finally statistics exports the colouring information of entire image.
5. method according to claim 2, is characterized in that, described sample image, when carrying out color moment feature extraction, first carries out piecemeal process to sample image.
6. method according to claim 2, is characterized in that, described sample image, when carrying out Corner Feature and extracting, carries out gray processing process to coloured image after resolution normalization operation terminates.
7. method according to claim 2, it is characterized in that, described sample image is when carrying out edge histogram feature extraction, after resolution normalization operation terminates, gaussian filtering is carried out to each color channel image in coloured image, picture breakdown is become to have overlapping image block, wherein every two adjacent image block Duplication are 0.3 to 0.5 times of its area, then the gradient distribution situation of statistical picture within the scope of 360 degree in each image block, finally statistics exports the marginal information description value of entire image.
8. method according to claim 1 and 2, it is characterized in that, after obtaining each category feature of Description Image, with Fuzzy C-Means Cluster Algorithm, clustering learning is carried out to sample characteristics, will merge to obtain overall training study parameter to selected sample characteristics data before training study.
9. method according to claim 1, is characterized in that, described in carry out initial results prediction steps comprise:
Obtain before initial results in calculating, carry out center traversal according to the breadth-first strategy of tree and calculate similarity, maximum two classes of last similarity are by as initial candidate result.
10. method according to claim 1, is characterized in that, describedly carries out learning training to often kind of books, obtains different training model of cognition steps comprise with this:
Adopt the learning algorithm based on Clustering, learning training is carried out to often kind of books, obtains different training model of cognition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510849994.9A CN105404682B (en) | 2015-06-12 | 2015-11-29 | A kind of book retrieval method based on digital image content |
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2015103258109 | 2015-06-12 | ||
| CN201510325810 | 2015-06-12 | ||
| CN201510849994.9A CN105404682B (en) | 2015-06-12 | 2015-11-29 | A kind of book retrieval method based on digital image content |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105404682A true CN105404682A (en) | 2016-03-16 |
| CN105404682B CN105404682B (en) | 2019-06-18 |
Family
ID=55470171
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510849994.9A Active CN105404682B (en) | 2015-06-12 | 2015-11-29 | A kind of book retrieval method based on digital image content |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105404682B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108288073A (en) * | 2018-01-30 | 2018-07-17 | 北京小米移动软件有限公司 | Picture authenticity identification method and device, computer readable storage medium |
| CN108921160A (en) * | 2018-05-04 | 2018-11-30 | 广东数相智能科技有限公司 | A kind of books recognition methods, electronic equipment and storage medium |
| WO2019210677A1 (en) * | 2018-05-04 | 2019-11-07 | Beijing Ling Technology Co., Ltd. | Method for Book Recognition and Book Reading Device |
| CN111695621A (en) * | 2020-06-09 | 2020-09-22 | 杭州印鸽科技有限公司 | System and method for detecting matching of customized content near-plane rule article and order based on deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103106202A (en) * | 2011-11-10 | 2013-05-15 | 周宝娟 | Book sharing system |
| CN103559314A (en) * | 2013-11-22 | 2014-02-05 | 孙生强 | Three-dimensional book data query system |
| US8694522B1 (en) * | 2012-03-28 | 2014-04-08 | Amazon Technologies, Inc. | Context dependent recognition |
| CN104391878A (en) * | 2014-10-31 | 2015-03-04 | 小米科技有限责任公司 | Book search method and book search device |
-
2015
- 2015-11-29 CN CN201510849994.9A patent/CN105404682B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103106202A (en) * | 2011-11-10 | 2013-05-15 | 周宝娟 | Book sharing system |
| US8694522B1 (en) * | 2012-03-28 | 2014-04-08 | Amazon Technologies, Inc. | Context dependent recognition |
| CN103559314A (en) * | 2013-11-22 | 2014-02-05 | 孙生强 | Three-dimensional book data query system |
| CN104391878A (en) * | 2014-10-31 | 2015-03-04 | 小米科技有限责任公司 | Book search method and book search device |
Non-Patent Citations (1)
| Title |
|---|
| 李宗民等: "九宫格空间框架的图书图像检索", 《中国图象图形学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108288073A (en) * | 2018-01-30 | 2018-07-17 | 北京小米移动软件有限公司 | Picture authenticity identification method and device, computer readable storage medium |
| CN108921160A (en) * | 2018-05-04 | 2018-11-30 | 广东数相智能科技有限公司 | A kind of books recognition methods, electronic equipment and storage medium |
| WO2019210677A1 (en) * | 2018-05-04 | 2019-11-07 | Beijing Ling Technology Co., Ltd. | Method for Book Recognition and Book Reading Device |
| CN111695621A (en) * | 2020-06-09 | 2020-09-22 | 杭州印鸽科技有限公司 | System and method for detecting matching of customized content near-plane rule article and order based on deep learning |
| CN111695621B (en) * | 2020-06-09 | 2023-05-05 | 杭州印鸽科技有限公司 | Method for detecting matching of customized article and order based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105404682B (en) | 2019-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107833213B (en) | Weak supervision object detection method based on false-true value self-adaptive method | |
| CN111444821A (en) | Automatic identification method for urban road signs | |
| CN108647649B (en) | Method for detecting abnormal behaviors in video | |
| CN108921083B (en) | Illegal mobile vendor identification method based on deep learning target detection | |
| CN111275082A (en) | Indoor object target detection method based on improved end-to-end neural network | |
| CN104615986B (en) | The method that pedestrian detection is carried out to the video image of scene changes using multi-detector | |
| CN110610166A (en) | Text region detection model training method and device, electronic equipment and storage medium | |
| CN109685065B (en) | Layout analysis method and system for automatically classifying test paper contents | |
| CN107341508B (en) | Fast food picture identification method and system | |
| CN106778650A (en) | Scene adaptive pedestrian detection method and system based on polymorphic type information fusion | |
| CN110781882A (en) | License plate positioning and identifying method based on YOLO model | |
| CN110991397B (en) | Travel direction determining method and related equipment | |
| CN105404682A (en) | Digital image content based book retrieval method | |
| CN111460927A (en) | Method for extracting structured information of house property certificate image | |
| CN107330027A (en) | A kind of Weakly supervised depth station caption detection method | |
| CN112749673A (en) | Method and device for intelligently extracting stock of oil storage tank based on remote sensing image | |
| CN111242899A (en) | Image-based flaw detection method and computer-readable storage medium | |
| CN108615401B (en) | Deep learning-based indoor non-uniform light parking space condition identification method | |
| CN113240623A (en) | Pavement disease detection method and device | |
| CN110737788B (en) | Rapid three-dimensional model index establishing and retrieving method | |
| CN111652117A (en) | Method and medium for segmenting multi-document image | |
| CN113505261B (en) | Data labeling method and device and data labeling model training method and device | |
| CN106548195A (en) | A kind of object detection method based on modified model HOG ULBP feature operators | |
| CN110889418A (en) | Gas contour identification method | |
| CN105160300A (en) | Text extraction method based on level set segmentation |
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 |