CN107944365A - A kind of system and method for Ceramic Cultural Relics intelligent recognition - Google Patents

A kind of system and method for Ceramic Cultural Relics intelligent recognition Download PDF

Info

Publication number
CN107944365A
CN107944365A CN201711133650.3A CN201711133650A CN107944365A CN 107944365 A CN107944365 A CN 107944365A CN 201711133650 A CN201711133650 A CN 201711133650A CN 107944365 A CN107944365 A CN 107944365A
Authority
CN
China
Prior art keywords
ceramic
cultural relics
feature
identification
genuine piece
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
Application number
CN201711133650.3A
Other languages
Chinese (zh)
Other versions
CN107944365B (en
Inventor
胡彩虹
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.)
In Kezhiwen (beijing) Technology Co Ltd
Original Assignee
In Kezhiwen (beijing) Technology Co Ltd
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 In Kezhiwen (beijing) Technology Co Ltd filed Critical In Kezhiwen (beijing) Technology Co Ltd
Priority to CN201711133650.3A priority Critical patent/CN107944365B/en
Publication of CN107944365A publication Critical patent/CN107944365A/en
Application granted granted Critical
Publication of CN107944365B publication Critical patent/CN107944365B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • 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
    • G06F16/51—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques

Abstract

The present invention relates to a kind of Ceramic Cultural Relics intelligent identifying system and method, based on the deep learning model of multidimensional characteristic matrix structure Ceramic Cultural Relics identification, solves the problems, such as that previous methods modeling accuracy is inadequate.Moreover, carrying out Ceramic Cultural Relics identification by the multidimensional characteristic storehouse and deep learning model for merging genuine piece storehouse, the operational efficiency of system is not only increased, and improves recognition effect, while self feed back update mechanism further ensures the accuracy of identification of this identifying system.

Description

A kind of system and method for Ceramic Cultural Relics intelligent recognition
Technical field：
The present invention relates to historical relic identification technology field, it particularly relates to towards ceramic-like historical relic picture feature multidimensional square Battle array and deep learning Model Identification technology, to ensure the demand of the accuracy of ceramics identification, real-time etc..
Background technology
Ceramic Cultural Relics are one of important carriers of our Chinese traditional cultures, are the treasure in China's culture and arts treasure-house Your legacy, is the valuable historical summary and wealth of the Chinese nation, therefore the correct identification of ceramics is particularly important.And make pottery at present The recognition methods of porcelain historical relic and means lack the foundation of system and science, and main authority Ceramic Cultural Relics differentiate expert in long-term mirror Not Shi Jian in some experiences for accumulating, these Heuristicses suffer from the limitation and identification person individual's quality and work bar in epoch The influence of part.At the same time manually differentiate also there are speed is slow, the cycle is long, poor reliability, it is costly the shortcomings of.Therefore, pattern is utilized The high-tech means such as identification theory, digital image processing techniques and computer deep learning build Ceramic Cultural Relics identifying system gesture It must go.
Existing many scholars are directed to the research work in terms of Ceramic Cultural Relics identification, and make some progress.Shanghai Silicate research institute develops one and applies mechanically the method that trace element carries out truth identification.But since trace element is Ceramic Cultural Relics The one side of discriminating so that its recognition result can only be as reference.Cambridge University develops thermohtminescence dating method, but its method Also simply from the age of single angle-determining Ceramic Cultural Relics, compared with traditional discrimination method, far away from traditional discrimination method science Comprehensively.
University of California at Irvine College of Computer Science in 2010 proposes a kind of based on the multi-class of layered weighting device Image partition method.This method defines hierarchical sequence based on object detector, and defined according to testing result it is right in scene The priori profile of elephant, then divide the image into as many super-pixel, with reference to profile priori, the hierarchical sequence of detector and different objects The color characteristic in region, builds unified frame, and declines two-step method using gradient and make inferences, and realizes more class objects in scene Semantic segmentation and hierarch recognition.Similar method also has the image scene structural analysis based on the study of Texton characteristic statistics Method and the picture structure analysis method based on regional model, it is main to handle object and its structural analysis in image scene, for Scene background structural analysis has certain limitation.Pu Chongliang (patent CN104361056A) proposed a kind of Gu in 2014 Ceramic true and false automatic identifying system and method, can utilize computer technology to complete the truth identification of ancient pottery and porcelain.But its method frame Structure and genuine piece property data base, collection apparatus and computer technology used in true and false identification module all have much room for improvement, with abundant The accuracy of identification and speed of Ceramic Cultural Relics are improved using existing information technology.
Analyzed based on more than, the present invention is directed to propose a kind of method and system of Ceramic Cultural Relics identification, this method and system The accuracy of identification of Ceramic Cultural Relics can be significantly improved, has especially also obtained large increase in extensive Ceramic Cultural Relics recognition speed.
The content of the invention
For drawbacks described above of the prior art, the present invention provides a kind of Ceramic Cultural Relics identifying system and method, specific skill Art scheme is as follows：
A kind of Ceramic Cultural Relics intelligent identifying system, including with lower part：
1) genuine piece multidimensional characteristic library module：Based on ceramic genuine piece picture library data, the structure side of feature multi-dimensional matrix is utilized Method, establishes genuine piece picture multidimensional characteristic storehouse；
2) Ceramic Cultural Relics characteristic extracting module to be identified：The image of Ceramic Cultural Relics to be identified is obtained, extracts ceramics to be identified The feature of historical relic simultaneously establishes its feature multi-dimensional matrix；
3) Ceramic Cultural Relics identification module：Genuine piece picture multidimensional characteristic storehouse and CRF deep learning models are merged, to pottery to be identified The microscopic features extraction result of porcelain historical relic carries out target identification, so as to obtain the recognition result of Ceramic Cultural Relics；
4) the self-built module in multidimensional characteristic storehouse：Rule is updated according to the renewal quantity set feature database of genuine piece picture library, once Triggering renewal rule, the self-built new genuine piece picture multidimensional characteristic storehouse of system off-line, and structure is completed into signal and feeds back to ceramic text Thing identification module；
5) Ceramic Cultural Relics identification model update module：Ceramic Cultural Relics identification module receives genuine piece picture multidimensional characteristic storehouse certainly After building the signal of completion, according to the renewal rule of setting again to genuine piece picture multidimensional characteristic storehouse and CRF deep learnings model into Row fusion, completes the renewal of Ceramic Cultural Relics identification model.
Further, the genuine piece picture multidimensional characteristic storehouse is to be based on genuine piece picture library, by traditional space co-occurrence matrix It is improved, mainly according to traditional pottery feature, the different gray features that carry out of shared weights calculate knot in ceramics identify Its textural characteristics is asked in the amendment of fruit, the calculating for being carried out at the same time Features of Fractal Dimension, so that the multidimensional for completing ceramic picture is special Levy the structure of matrix.
Further, genuine piece ceramics picture PxMultidimensional characteristic be respectively：Moon bottle is embraced, celadon, under-glaze red, plain tricolour, goes through History personage, folding branch flower, decorative pattern poetic atmosphere, check design, according to traditional gray level co-occurrence matrixes processing method, calculates PxEight sides herein The characteristic value in face, is respectively Px{A1, A2, A3, A4, A5, A6, A7, A8, by this characteristic value result and each feature in ceramics identify Shared weights are multiplied, setting genuine piece ceramics picture PxGray feature value correction value be Px{A11, A22, A33, A44, A55, A66, A77, A88}；
Be then based on this gray scale vector and carry out dimension calculating, the Features of Fractal Dimension vector D obtained after calculating and gray scale to Amount collectively forms genuine piece ceramics picture PxMultidimensional characteristic vectors { D, A11, A22, A33, A44, A55, A66, A77, A88}；
Step handles all ceramic genuine piece pictures according to this, finally obtains the multidimensional characteristic square in ceramic genuine piece storehouse Battle array.
Further, when carrying out dimension calculating based on gray scale vector, window subgraph size value is 32 × 32 pictures Element.
Further, the CRF deep learnings model construction is based on the Target Recognition Algorithms CRF towards ceramics, that is, is existed With reference to the feature set scope and identification level of traditional ceramics identification on the basis of classical Target Recognition Algorithms Fast-RCNN, change The characteristic pattern amount of area and the process of convolution number of plies of Fast-RCNN.
Further, pottery feature weights are added in the Feature Mapping of different levels, obtain the prediction knot of different specific weight Fruit；Meanwhile extra convolutional layer is added in the infrastructure network of Fast-RCNN, the size of the convolutional layer is successively to pass Subtract, can be calculated in multiple dimensioned lower progress object prediction；During the CRF deep learnings model training, the spy in training image Value indicative is imparted on the bounding box of fixed output；Model output is that predefined is good, a series of bounding box of fixed sizes And its score s of relatively a certain featurek, score calculation formula is as follows：
Wherein sminAnd smaxIt is the score minimum value and maximum for being fitted, m is the set sizes of Feature Mapping.
Further, the recognition methods of the Ceramic Cultural Relics identification module is：First by ceramic image multi-dimensional to be identified The data in eigenmatrix and genuine piece multidimensional characteristic storehouse carry out similarity measure, the feature vector obtained after ceramic image acquisition and processing For Q, first Q and each feature vector P in genuine piece storehousexSimilarity measure is carried out, calculation formula is：
If the similarity of result of calculation is not above the genuine piece similar threshold value of setting, end of identification, it is similar to provide identification Degree；If the similarity of result of calculation exceedes the genuine piece similar threshold value of setting, further by ceramic feature square to be identified Battle array is input to CRF deep learnings model and carries out target identification, multi-feature recognition result is carried out to integrate output, i.e.,：Will be to be identified The feature vector Q of ceramic picture picture is input to CRF models and marking is identified, the specific category according to belonging to marking result obtains Q.
Further, the update mechanism of the Ceramic Cultural Relics identification model update module is：When the genuine piece picture library more When new quantity reaches the threshold value of setting, the self-built new multidimensional characteristic storehouse of system off-line, and structure is completed into signal transmission to ceramics Historical relic identification module.
One kind is also provided Ceramic Cultural Relics intelligent identification Method is carried out using above-mentioned Ceramic Cultural Relics intelligent identifying system, including such as Lower step：
Step 1：Ceramic Cultural Relics feature multi-dimensional matrix is built, ceramic picture is gathered from multiple dimensions, to Image Acquisition Result carry out gray feature calculating first with the space co-occurrence matrix of modified adaptive weight, it is special then to carry out dimension Its textural characteristics is asked in the calculating of sign, completes the structure of the multidimensional characteristic matrix of ceramic picture；
Step 2：Deep learning model is established, the multidimensional characteristic matrix based on ceramic picture, foundation is based on object The ceramic-like historical relic Target Recognition Algorithms CRF of detection algorithm Fast-RCNN, and the deep learning towards ceramics identification is built according to this Model, realizes the identification of Ceramic Cultural Relics.
Further, multiple dimensions include：
1) type：
Bottle：Plum vase, garlic bottle, beautiful pot spring bottle, vault-of-Heaven vase, armful moon bottle, flask etc.；Honor：Ring a bell honor, without shelves honor, horseshoe Honor, too white honor；Pot：Chicken head pots, bale handle pot, flat pot, hold pot；Bowl：Heart bowl, pier formula bowl, bowl of bowing, Gao Zuwan；Cup.
2) technique：
Color vitreous enamel：Celadon, ceramic whiteware, black porcelain；Underglaze colour：Blue and white, under-glaze red, blue and white with copper red colors：Doucai contrasting colors；Overglaze color：Five Coloured silk, powder enamel, sail are red, spearmint, plain tricolour.
3) decorative pattern：
Personage's line：Historical personage, Buddhism personage, Taoism personage, opera personage, baby；Animal line：Auspicious beast, natural animal； Plant line：Twine a line, folding branch flower, lucky, decorative pattern poetic atmosphere；Geometry line：Bright and beautiful ground-tint, raised line design, reticulate pattern, check design, connection pearl line；Word Line；Have good luck line.
The technique effect of the present invention, Ceramic Cultural Relics recognition methods provided by the invention and system, based on multidimensional characteristic matrix The deep learning model of Ceramic Cultural Relics identification is built, solves the problems, such as that previous methods modeling accuracy is inadequate.Moreover, pass through fusion The multidimensional characteristic storehouse in genuine piece storehouse and deep learning model carry out Ceramic Cultural Relics identification, not only increase the operational efficiency of system, and And recognition effect is improved, while self feed back update mechanism further ensures the accuracy of identification of this identifying system.
Brief description of the drawings
Fig. 1 is the online process chart of Ceramic Cultural Relics intelligent identifying system of the present invention；
Fig. 2 is CRF deep learnings model construction schematic diagram of the present invention；
Fig. 3 is the identification process flow diagram of the ceramic identification module of the present invention；
Fig. 4 is the self feed back update mechanism schematic diagram of Ceramic Cultural Relics identifying system of the present invention；
Embodiment
In order that those skilled in the art will better understand the technical solution of the present invention, and make the present invention above-mentioned mesh , feature and advantage can be more obvious understandable, with reference to embodiment attached drawing, the present invention is described in further detail.This Field technology personnel it is clear that, can also realize this hair in the case where lacking these part or all of details It is bright.In other cases, in order not to the present invention can be made not specifically describe known processing there are unnecessary unclear part Step and/or structure.In addition, although present invention is described in conjunction with the specific embodiments, it should be understood that, this is retouched State to be not intended as and invention is limited to described embodiment.It may include to want by appended right on the contrary, the description is intended to covering Seek replacement, improvement and the equivalent in the spirit and scope of the present invention of book restriction.
With reference to figure 1, the present invention provides a kind of Ceramic Cultural Relics identifying system, online process flow such as Fig. 1 institutes of the system Show.The structure in genuine piece multidimensional characteristic storehouse is carried out first, is next based on Target Recognition Algorithms CRF structures towards the depth of ceramics identification Learning model, forms ceramic identifying system on this basis.During application on site, by the multidimensional characteristic square of ceramic image to be identified Battle array submits to ceramic identifying system as input information, and then obtains final ceramic recognition result.
Specific to each step：Genuine piece multidimensional characteristic storehouse structure is to be based on genuine piece picture library, by traditional space symbiosis Matrix is improved, and mainly according to traditional pottery feature, the different of shared weights carry out gray feature meter in ceramics identify The amendment of result is calculated, its textural characteristics is asked in the calculating for being carried out at the same time Features of Fractal Dimension, so as to complete the more of ceramic picture The structure of dimensional feature matrix.Assuming that the characteristic set built for ceramic picture as us includes eight aspects, genuine piece ceramics picture PxMultidimensional characteristic be respectively embrace moon bottle, celadon, under-glaze red, plain tricolour, historical personage, folding branch flower, decorative pattern poetic atmosphere, grid Line }, according to traditional gray level co-occurrence matrixes processing method, calculate PxThe characteristic value of eight aspects herein, is respectively Px{A1, A2, A3, A4, A5, A6, A7, A8, shared weights are multiplied during in the present invention, we identify this characteristic value result with each feature in ceramics, Set genuine piece ceramics picture PxGray feature value correction value be Px{A11, A22, A33, A44, A55, A66, A77, A88}.It is then based on This gray scale vector carries out dimension calculating, needs to consider the selection of subgraph window at this time and calculates the selection of dimension scale, Avoid because the too small important textural characteristics or window subgraph size lost of window subgraph size cause greatly very much edge pixel Mixed with other pixels of image-region and influence the selection of textural characteristics.Window subgraph size value 32 in the present embodiment × 32 pixels.The Features of Fractal Dimension vector D obtained after calculating collectively forms genuine piece ceramics picture P with gray scale vectorxMultidimensional characteristic Vector { D, A11, A22, A33, A44, A55, A66, A77, A88}.Step handles all ceramic genuine piece pictures according to this, finally Obtain the multidimensional characteristic matrix in ceramic genuine piece storehouse.
CRF deep learning model constructions are based on the Target Recognition Algorithms CRF towards ceramics.CRF algorithms are in classics With reference to the feature set scope and identification level of traditional ceramics identification on the basis of Target Recognition Algorithms Fast-RCNN, change Fast- The characteristic pattern amount of area and the process of convolution number of plies of RCNN, as shown in Figure 2.The core of CRF methods is exactly to predict object and its ownership The score of classification；Small convolution kernel is used in Feature Mapping at the same time, goes to predict a series of bounding boxes and boundary shifts amount.In order to The precision of high ceramics identification is obtained, pottery feature weights are added in the Feature Mapping of different levels, obtain the pre- of different specific weight Survey result.In addition, with the addition of extra convolutional layer in the infrastructure network of Fast-RCNN, the size of these convolutional layers is Successively successively decrease, can be calculated in multiple dimensioned lower progress object prediction.During CRF deep learning model trainings, in training image Characteristic value needs to be imparted on the bounding box of those fixation outputs.Model output is that predefined is good, and a series of fixations are big Small bounding box and its score s of relatively a certain featurek, score calculation formula is as follows：
Wherein sminAnd smaxIt is the score minimum value and maximum for being fitted, m is the set sizes of Feature Mapping.This reality Apply s in exampleminAnd smaxValue 0.01 and the value of 0.99, m are 8.
Ceramic identification module is then to have merged genuine piece multidimensional characteristic storehouse and utilized the trained CRF in genuine piece multidimensional characteristic storehouse deep Spend learning model and carry out corresponding identifying processing.First by ceramic picture to be identified as multidimensional characteristic matrix and genuine piece multidimensional characteristic The data in storehouse carry out similarity measure, if the similarity of result of calculation is not above the genuine piece similar threshold value of setting, identification knot Beam, provides identification similarity；If the similarity of result of calculation exceedes the genuine piece similar threshold value of setting, will further wait to know Ceramic eigenmatrix is not input to CRF deep learnings model and carries out target identification, multi-feature recognition result integrate defeated Go out, identification process is as shown in Figure 3.In the present embodiment after ceramics image acquisition and processing to be identified obtained feature vector for Q E, B11, B22, B33, B44, B55, B66, B77, B88}.Q first and each feature vector P in genuine piece storehousexCarry out similarity measure, calculation formula For：When some similarity measure result be more than 0.8 when, by the feature of ceramic picture picture to be identified to Amount Q is input to CRF models and marking is identified, the specific category according to belonging to marking result obtains Q.
The self feed back update mechanism of Ceramic Cultural Relics identifying system is as shown in Figure 4.Set when genuine piece picture library renewal quantity reaches During fixed threshold value, the self-built new multidimensional characteristic storehouse of system off-line, and structure is completed into signal transmission and gives Ceramic Cultural Relics identification module. After Ceramic Cultural Relics identification module receives the signal of the self-built completion in multidimensional characteristic storehouse, according to the renewal rule of setting again to multidimensional Feature database and deep learning model are merged, and complete the renewal of Ceramic Cultural Relics identification model.
The present invention is exemplarily described above, it is clear that present invention specific implementation is not subject to the restrictions described above, As long as employing the various improvement of inventive concept and technical scheme of the present invention progress, or not improved directly apply to other fields Close, within protection scope of the present invention.

Claims (10)

1. a kind of Ceramic Cultural Relics intelligent identifying system, it is characterised in that including with lower part：
1) genuine piece multidimensional characteristic library module：Based on ceramic genuine piece picture library data, using the construction method of feature multi-dimensional matrix, build Vertical genuine piece picture multidimensional characteristic storehouse；
2) Ceramic Cultural Relics characteristic extracting module to be identified：The image of Ceramic Cultural Relics to be identified is obtained, extracts Ceramic Cultural Relics to be identified Feature and establish its feature multi-dimensional matrix；
3) Ceramic Cultural Relics identification module：Genuine piece picture multidimensional characteristic storehouse and CRF deep learning models are merged, to ceramics text to be identified The microscopic features extraction result of thing carries out target identification, so as to obtain the recognition result of Ceramic Cultural Relics；
4) the self-built module in multidimensional characteristic storehouse：Rule is updated according to the renewal quantity set feature database of genuine piece picture library, once triggering Renewal rule, the self-built new genuine piece picture multidimensional characteristic storehouse of system off-line, and structure is completed into signal and feeds back to Ceramic Cultural Relics knowledge Other module；
5) Ceramic Cultural Relics identification model update module：It is self-built complete that Ceramic Cultural Relics identification module receives genuine piece picture multidimensional characteristic storehouse Into signal after, genuine piece picture multidimensional characteristic storehouse and CRF deep learning models are melted again according to the renewal rule of setting Close, complete the renewal of Ceramic Cultural Relics identification model.
2. Ceramic Cultural Relics intelligent identifying system as claimed in claim 1, it is characterised in that：
The genuine piece picture multidimensional characteristic storehouse is to be based on genuine piece picture library, traditional space co-occurrence matrix is improved, mainly It is that the difference of shared weights carries out the amendments of gray feature result of calculations in ceramics identify according to traditional pottery feature, at the same time Its textural characteristics is asked in the calculating for carrying out Features of Fractal Dimension, so as to complete the structure of the multidimensional characteristic matrix of ceramic picture.
3. Ceramic Cultural Relics intelligent identifying system as claimed in claim 2, it is characterised in that：
Genuine piece ceramics picture PxMultidimensional characteristic be respectively：Embrace moon bottle, celadon, under-glaze red, plain tricolour, historical personage, folding branch flower, Decorative pattern poetic atmosphere, check design, according to traditional gray level co-occurrence matrixes processing method, calculates PxThe characteristic value of eight aspects herein, point Wei not Px{A1, A2, A3, A4, A5, A6, A7, A8, by this characteristic value result, the shared weights in ceramics identify are multiplied with each feature, Set genuine piece ceramics picture PxGray feature value correction value be Px{A11, A22, A33, A44, A55, A66, A77, A88}；
It is then based on this gray scale vector and carries out dimension calculating, the Features of Fractal Dimension vector D obtained after calculating and gray scale vector is common Isomorphism is into genuine piece ceramic picture piece PxMultidimensional characteristic vectors { D, A11, A22, A33, A44, A55, A66, A77, A88}；
Step handles all ceramic genuine piece pictures according to this, finally obtains the multidimensional characteristic matrix in ceramic genuine piece storehouse.
4. Ceramic Cultural Relics intelligent identifying system as claimed in claim 3, it is characterised in that：
When carrying out dimension calculating based on gray scale vector, window subgraph size value is 32 × 32 pixels.
5. Ceramic Cultural Relics intelligent identifying system as claimed in claim 1, it is characterised in that：
The CRF deep learnings model construction is based on the Target Recognition Algorithms CRF towards ceramics, i.e., is calculated in classical target identification The feature set scope and identification level identified on the basis of method Fast-RCNN with reference to traditional ceramics, changes the feature of Fast-RCNN Graph region amount and the process of convolution number of plies.
6. Ceramic Cultural Relics intelligent identifying system as claimed in claim 5, it is characterised in that：
Pottery feature weights are added in the Feature Mapping of different levels, obtain the prediction result of different specific weight；Meanwhile Extra convolutional layer is added in the infrastructure network of Fast-RCNN, the size of the convolutional layer is successively successively decreased, Ke Yi Multiple dimensioned lower progress object prediction calculates；During the CRF deep learnings model training, the characteristic value in training image is imparted to solid Surely on the bounding box exported；Model output is that predefined is good, a series of bounding box of fixed sizes and its relatively a certain The score s of featurek, score calculation formula is as follows：
<mrow> <msub> <mi>s</mi> <mi>k</mi> </msub> <mo>=</mo> <msub> <mi>s</mi> <mi>min</mi> </msub> <mo>+</mo> <mfrac> <mrow> <msub> <mi>s</mi> <mi>max</mi> </msub> <mo>-</mo> <msub> <mi>s</mi> <mi>min</mi> </msub> </mrow> <mrow> <mi>m</mi> <mo>-</mo> <mn>1</mn> </mrow> </mfrac> <mrow> <mo>(</mo> <mi>k</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>,</mo> <mi>k</mi> <mo>&amp;Element;</mo> <mo>&amp;lsqb;</mo> <mn>1</mn> <mo>,</mo> <mi>m</mi> <mo>&amp;rsqb;</mo> </mrow>
Wherein sminAnd smaxIt is the score minimum value and maximum for being fitted, m is the set sizes of Feature Mapping.
7. Ceramic Cultural Relics intelligent identifying system as claimed in claim 1, it is characterised in that：
The recognition methods of the Ceramic Cultural Relics identification module is：First by ceramic picture to be identified as multidimensional characteristic matrix and genuine piece The data in multidimensional characteristic storehouse carry out similarity measure, and the feature vector obtained after ceramic image acquisition and processing be Q, Q first with very Each feature vector P in product storehousexSimilarity measure is carried out, calculation formula is：
If the similarity of result of calculation is not above the genuine piece similar threshold value of setting, end of identification, provides identification similarity i.e. Can；If the similarity of result of calculation exceedes the genuine piece similar threshold value of setting, further that ceramic eigenmatrix to be identified is defeated Enter to CRF deep learnings model and carry out target identification, multi-feature recognition result is carried out to integrate output, i.e.,：By ceramics to be identified The feature vector Q of image is input to CRF models and marking is identified, the specific category according to belonging to marking result obtains Q.
8. Ceramic Cultural Relics intelligent identifying system as claimed in claim 1, it is characterised in that：
The update mechanism of the Ceramic Cultural Relics identification model update module is：When genuine piece picture library renewal quantity reaches setting Threshold value when, the self-built new multidimensional characteristic storehouse of system off-line, and will structure complete signal transmission give Ceramic Cultural Relics identification module.
9. a kind of carry out Ceramic Cultural Relics intelligent recognition side using such as Ceramic Cultural Relics intelligent identifying system any claim 1-8 Method, includes the following steps：
Step 1：Ceramic Cultural Relics feature multi-dimensional matrix is built, ceramic picture is gathered from multiple dimensions, to the knot of Image Acquisition Fruit carries out gray feature calculating first with the space co-occurrence matrix of modified adaptive weight, then carries out Features of Fractal Dimension Its textural characteristics is asked in calculating, completes the structure of the multidimensional characteristic matrix of ceramic picture；
Step 2：Deep learning model is established, the multidimensional characteristic matrix based on ceramic picture, foundation is based on target analyte detection The ceramic-like historical relic Target Recognition Algorithms CRF of algorithm Fast-RCNN, and the deep learning mould towards ceramics identification is built according to this Type, realizes the identification of Ceramic Cultural Relics.
10. Ceramic Cultural Relics intelligent identification Method as claimed in claim 9, it is characterised in that the multiple dimension includes：
1) type：
Bottle：Plum vase, garlic bottle, beautiful pot spring bottle, vault-of-Heaven vase, armful moon bottle, flask etc.；Honor：Ring a bell honor, without shelves honor, horseshoe honor, too White honor；Pot：Chicken head pots, bale handle pot, flat pot, hold pot；Bowl：Heart bowl, pier formula bowl, bowl of bowing, Gao Zuwan；Cup.
2) technique：
Color vitreous enamel：Celadon, ceramic whiteware, black porcelain；Underglaze colour：Blue and white, under-glaze red, blue and white with copper red colors：Doucai contrasting colors；Overglaze color：The five colours, powder Coloured silk, sail are red, spearmint, plain tricolour.
3) decorative pattern：
Personage's line：Historical personage, Buddhism personage, Taoism personage, opera personage, baby；Animal line：Auspicious beast, natural animal；Plant Line：Twine a line, folding branch flower, lucky, decorative pattern poetic atmosphere；Geometry line：Bright and beautiful ground-tint, raised line design, reticulate pattern, check design, connection pearl line；Word line； Have good luck line.
CN201711133650.3A 2017-11-16 2017-11-16 Intelligent identification system and method for ceramic cultural relics Active CN107944365B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711133650.3A CN107944365B (en) 2017-11-16 2017-11-16 Intelligent identification system and method for ceramic cultural relics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711133650.3A CN107944365B (en) 2017-11-16 2017-11-16 Intelligent identification system and method for ceramic cultural relics

Publications (2)

Publication Number Publication Date
CN107944365A true CN107944365A (en) 2018-04-20
CN107944365B CN107944365B (en) 2020-08-11

Family

ID=61932477

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711133650.3A Active CN107944365B (en) 2017-11-16 2017-11-16 Intelligent identification system and method for ceramic cultural relics

Country Status (1)

Country Link
CN (1) CN107944365B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110942076A (en) * 2019-11-27 2020-03-31 清华大学 Method and system for generating anti-counterfeiting mark of ceramic product
CN111160145A (en) * 2019-12-13 2020-05-15 广州懿锝文化创意有限公司 Method for identifying uniqueness of porcelain
CN112101448A (en) * 2020-09-10 2020-12-18 敬科(深圳)机器人科技有限公司 Screen image recognition method, device and system and readable storage medium

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101799292A (en) * 2010-02-25 2010-08-11 马建春 Relic identification method
CN103116755A (en) * 2013-01-27 2013-05-22 深圳市书圣艺术品防伪鉴定有限公司 Automatic painting and calligraphy authenticity degree detecting system and method thereof
CN104361056A (en) * 2014-10-30 2015-02-18 蒲重良 Automatic ancient ceramic authenticity identification system and method
CN104504360A (en) * 2014-09-28 2015-04-08 中国人民公安大学 Automatic authentication method for ancient ceramics
CN106096028A (en) * 2016-06-24 2016-11-09 北京旷视科技有限公司 Historical relic indexing means based on image recognition and device
CN106408037A (en) * 2015-07-30 2017-02-15 阿里巴巴集团控股有限公司 Image recognition method and apparatus
CN107346432A (en) * 2016-05-04 2017-11-14 上海优尼客物联网有限公司 Feature extraction applied to the enhancing of ware

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101799292A (en) * 2010-02-25 2010-08-11 马建春 Relic identification method
CN103116755A (en) * 2013-01-27 2013-05-22 深圳市书圣艺术品防伪鉴定有限公司 Automatic painting and calligraphy authenticity degree detecting system and method thereof
CN104504360A (en) * 2014-09-28 2015-04-08 中国人民公安大学 Automatic authentication method for ancient ceramics
CN104361056A (en) * 2014-10-30 2015-02-18 蒲重良 Automatic ancient ceramic authenticity identification system and method
CN106408037A (en) * 2015-07-30 2017-02-15 阿里巴巴集团控股有限公司 Image recognition method and apparatus
CN107346432A (en) * 2016-05-04 2017-11-14 上海优尼客物联网有限公司 Feature extraction applied to the enhancing of ware
CN106096028A (en) * 2016-06-24 2016-11-09 北京旷视科技有限公司 Historical relic indexing means based on image recognition and device

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110942076A (en) * 2019-11-27 2020-03-31 清华大学 Method and system for generating anti-counterfeiting mark of ceramic product
CN110942076B (en) * 2019-11-27 2020-10-16 清华大学 Method and system for generating anti-counterfeiting mark of ceramic product
CN111160145A (en) * 2019-12-13 2020-05-15 广州懿锝文化创意有限公司 Method for identifying uniqueness of porcelain
CN111160145B (en) * 2019-12-13 2023-09-19 广州懿锝文化创意有限公司 Method for identifying porcelain uniqueness
CN112101448A (en) * 2020-09-10 2020-12-18 敬科(深圳)机器人科技有限公司 Screen image recognition method, device and system and readable storage medium

Also Published As

Publication number Publication date
CN107944365B (en) 2020-08-11

Similar Documents

Publication Publication Date Title
CN108717568B (en) A kind of image characteristics extraction and training method based on Three dimensional convolution neural network
CN105787439B (en) A kind of depth image human synovial localization method based on convolutional neural networks
CN104217214B (en) RGB D personage&#39;s Activity recognition methods based on configurable convolutional neural networks
CN108171266A (en) A kind of learning method of multiple target depth convolution production confrontation network model
CN109101930A (en) A kind of people counting method and system
CN107133960A (en) Image crack dividing method based on depth convolutional neural networks
CN110084318A (en) A kind of image-recognizing method of combination convolutional neural networks and gradient boosted tree
CN109447078A (en) A kind of detection recognition method of natural scene image sensitivity text
CN103761536B (en) Human face beautifying method based on non-supervision optimal beauty features and depth evaluation model
CN108875934A (en) A kind of training method of neural network, device, system and storage medium
CN108182441A (en) Parallel multichannel convolutive neural network, construction method and image characteristic extracting method
CN108960499A (en) A kind of Fashion trend predicting system merging vision and non-vision feature
CN107944365A (en) A kind of system and method for Ceramic Cultural Relics intelligent recognition
CN106909924A (en) A kind of remote sensing image method for quickly retrieving based on depth conspicuousness
CN110533737A (en) The method generated based on structure guidance Chinese character style
CN108830913A (en) Semantic level line original text painting methods based on User Colors guidance
CN109492666A (en) Image recognition model training method, device and storage medium
CN109684912A (en) A kind of video presentation method and system based on information loss function
CN107358293A (en) A kind of neural network training method and device
CN112070078B (en) Deep learning-based land utilization classification method and system
CN109063724A (en) A kind of enhanced production confrontation network and target sample recognition methods
CN109271895A (en) Pedestrian&#39;s recognition methods again based on Analysis On Multi-scale Features study and Image Segmentation Methods Based on Features
CN106599805A (en) Supervised data driving-based monocular video depth estimating method
CN106529503A (en) Method for recognizing face emotion by using integrated convolutional neural network
CN109146989A (en) A method of birds and flowers characters in a fancy style image is generated by building neural network

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
DD01 Delivery of document by public notice

Addressee: Zhongke Zhiwen (Beijing) Technology Co., Ltd.

Document name: the First Notification of an Office Action

DD01 Delivery of document by public notice
GR01 Patent grant
GR01 Patent grant