CN109583376A - The disconnected source periodization method of ancient pottery and porcelain based on multicharacteristic information fusion - Google Patents

The disconnected source periodization method of ancient pottery and porcelain based on multicharacteristic information fusion Download PDF

Info

Publication number
CN109583376A
CN109583376A CN201811451703.0A CN201811451703A CN109583376A CN 109583376 A CN109583376 A CN 109583376A CN 201811451703 A CN201811451703 A CN 201811451703A CN 109583376 A CN109583376 A CN 109583376A
Authority
CN
China
Prior art keywords
porcelain
image
ancient pottery
ancient
pottery
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
CN201811451703.0A
Other languages
Chinese (zh)
Other versions
CN109583376B (en
Inventor
周强
张静
王莹
张瑞瑞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shaanxi University of Science and Technology
Original Assignee
Shaanxi University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shaanxi University of Science and Technology filed Critical Shaanxi University of Science and Technology
Priority to CN201811451703.0A priority Critical patent/CN109583376B/en
Publication of CN109583376A publication Critical patent/CN109583376A/en
Application granted granted Critical
Publication of CN109583376B publication Critical patent/CN109583376B/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
    • G06V20/80—Recognising image objects characterised by unique random patterns
    • 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
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The present invention relates to a kind of disconnected source periodization methods of the ancient pottery and porcelain based on multicharacteristic information fusion, ancient pottery and porcelain implements image is extracted from ancient pottery and porcelain image realizes background separation, it is partitioned into accurate ancient pottery and porcelain image, then denoising is filtered to ancient pottery and porcelain image and obtains clearly image；From two type in image, color feature spaces while type structure feature, glaze color characteristic are extracted, be input to the completion of SAE-DBN network to the information fusion of the characteristic quantity of the ancient pottery and porcelain image and the layer-by-layer extraction of characteristic quantity as characteristic quantity and is abstracted；Ancient pottery and porcelain image to be measured is input to trained DBN classifier to be trained, the classification of ancient pottery and porcelain is completed by training result and the disconnected source of ancient pottery and porcelain, the division of history into periods identify.The present invention is able to achieve classification and the disconnected source, division of history into periods identification of ancient pottery and porcelain, provides objective, reliable, quantization, accurate method for the disconnected source division of history into periods identification of ancient pottery and porcelain.

Description

The disconnected source periodization method of ancient pottery and porcelain based on multicharacteristic information fusion
Technical field
The invention belongs to ancient pottery and porcelain historical relics to identify field, and in particular to a kind of ancient pottery and porcelain based on multicharacteristic information fusion is disconnected Source periodization method.
Background technique
The high reserve value of ancient pottery and porcelain and auction price cause the imitative product largely mixed the spurious with the genuine to occur, and bring to society huge Big economic loss, so the accurate disconnected source division of history into periods identification to ancient pottery and porcelain historical relic is particularly important.
Currently, mainly have traditional experience identification method and modern science and technology identification method for ancient pottery and porcelain identification method, One by virtue of experience identifies shortage science and confidence level；Secondly being deposited using the method that modern means of science and technology are identified to ancient pottery and porcelain sample In a degree of destructiveness, and the elemental composition of fakement altitude simulation ancient pottery and porcelain historical relic, thermoluminescence spectrum etc. will lead to scientific instrument Device has the case where erroneous judgement, therefore for the disconnected source division of history into periods identification of ancient pottery and porcelain, there are significant limitations at present.
Summary of the invention
The object of the present invention is to provide a kind of disconnected source periodization methods of the ancient pottery and porcelain based on multicharacteristic information fusion, may be implemented Its disconnected source division of history into periods detection is completed according to a width antique pottery porcelain photograph.
The technical scheme adopted by the invention is as follows:
The disconnected source periodization method of ancient pottery and porcelain based on multicharacteristic information fusion, it is characterised in that:
Ancient pottery and porcelain implements image is extracted from ancient pottery and porcelain image and realizes background separation, is partitioned into accurate ancient pottery and porcelain image, Denoising is filtered to ancient pottery and porcelain image again and obtains clearly image；
Type structure feature, glaze color characteristic are extracted simultaneously from two type in image, color feature spaces, as feature Amount is input to information fusion and the layer-by-layer extraction of characteristic quantity of the SAE-DBN network completion to the characteristic quantity of the ancient pottery and porcelain image With it is abstract；
Ancient pottery and porcelain image to be measured is input to trained DBN classifier to be trained, antique pottery is completed by training result The classification of porcelain and the disconnected source of ancient pottery and porcelain, the division of history into periods identify.
The method includes preprocessing module, characteristic extracting module, the disconnected source division of history into periods modules of ancient pottery and porcelain；
It is clear to obtain that the preprocessing module uses secondary treatment method to carry out prospect background separation to ancient pottery and porcelain image Ancient pottery and porcelain image and send to characteristic extracting module；
Characteristic extracting module is used to extract antique pottery porcelain simultaneously in two feature spaces of type, glaze colours of ancient pottery and porcelain image Two kinds of type, glaze colours characteristic quantities；
The ancient pottery and porcelain is broken source division of history into periods module, is to complete the spy to the ancient pottery and porcelain image using SAE-DBN network model The information of sign amount merges, and by the layer-by-layer extraction to characteristic quantity and is abstracted, may finally realize classification and the Gu of ancient pottery and porcelain image The disconnected source division of history into periods of ceramics identifies.
The preprocessing module use information second extraction method, including processing for the first time and second of processing, for real The prospect background of existing ancient pottery and porcelain image is separated to obtain clearly ancient pottery and porcelain target object image；
The first time processing is utilization " pyramid " operator to tri- colors of RGB point of each pixel of ancient pottery and porcelain image Amount carries out convolution sum processing respectively, then chooses appropriate threshold and extract edge, obtains a first colour edging profile i.e. edge graph Picture；
Second of processing is that ancient pottery and porcelain display foreground background separation is realized by mask matching method, is calculated first each Edge image is transformed into gray level image by the weighted grey-value of tri- color components of RGB of pixel, then passes through Value-mean filter method eliminates the noise jamming in image, carries out characteristic information extraction process, that is, second edge extraction process again, mentions It gets the complete edge binary images of ancient pottery and porcelain and is known as ancient pottery and porcelain mask images, then ancient pottery and porcelain mask images and original image are compound Operation, can be in the ancient pottery and porcelain image that former background and antique pottery porcelain body are difficult to differentiate between before the realization of complete extraction ancient pottery and porcelain implements region Scape background separation.
The extraction ancient ceramics structure information process is by profile extraction module, main view correction module, profile function Fitting module three parts image processing operations composition；
The profile extraction module is at the ancient pottery and porcelain image denoising after extraction prospect background separation and binaryzation Reason, then edge contour is extracted, including top, foot ellipse, outline line on ancient pottery and porcelain implements；
The ancient pottery and porcelain image border profile extracted according to the main view correction module method determines that ancient pottery and porcelain is shot Angle is restored to front view, i.e. ancient pottery and porcelain front view to ancient pottery and porcelain image rectification；
The functional characteristics that the profile function fitting module is approached using BP neural network height is to ancient pottery and porcelain front view weight It is new to carry out edge contour extraction and Function Fitting；The neural network needs to choose optimum network structural parameters, by training The weight and threshold value between each node are obtained, the optimum fit curve for meeting required precision is obtained, is obtained by ancient pottery and porcelain contour curve Multiple type the amount of characteristic parameter of ancient pottery and porcelain: inflection point number n, area perimeter fractal box fac, the height of neck/stomach heightAncient pottery and porcelain foot area/upper top areaReconstruct three-dimensional ancient pottery and porcelain body surface area/ancient pottery and porcelain volumeIt is generated One group of assemblage characteristic vector is to characterize ancient ceramics structure feature.
The extraction of the ancient pottery and porcelain glaze colours information, first to improve ancient pottery and porcelain image segmentation at several pockets Accuracy in computation, then ancient pottery and porcelain RGB color is transformed into hsv color space, and by H, the S in hsv color space two Color interval is divided into several small color interval i.e. color levels, calculates two face of H, S of each ancient pottery and porcelain image fritter Pixel quantity of the colouring component in each color level obtains color histogram, and horizontal axis indicates color level range, and the longitudinal axis indicates picture Prime number amount；To the color component feature of each segmented image block in whole picture ancient pottery and porcelain image, weighted average is handled respectively again, is obtained The color histogram feature of entire image calculates the color assemblage characteristic vector of its comentropy and energy as ancient pottery and porcelain image, That is ancient pottery and porcelain glaze colours information.
The ancient pottery and porcelain is broken source division of history into periods module, is broken source division of history into periods detection using ancient pottery and porcelain is carried out based on SAE-DBN network model； Ancient ceramics structure feature is extracted simultaneously respectively from ancient pottery and porcelain type feature space and color feature space and color characteristic is believed Breath, is generated assemblage characteristic vector, is broken the judgment basis of the source division of history into periods as ancient pottery and porcelain；The characteristic quantity extracted is inputted To SAE-DBN network, completes to merge the information of the characteristic quantity of the ancient pottery and porcelain image, information is melted using sparse self-encoding encoder The characteristic quantity of conjunction carries out dimension-reduction treatment to obtain better feature description, then characteristic is input in depth confidence network and is instructed Practice DBN classifier, the classification of ancient pottery and porcelain image to be measured is completed according to training result and disconnected source, the division of history into periods identify.
The concrete methods of realizing of the disconnected source division of history into periods of ancient pottery and porcelain is to extract using the ancient pottery and porcelain image in database as training sample To the feature vector of training sample be input to SAE-DBN network and complete network training and to generate trained DBN classifier, then Using antique pottery porcelain photograph to be measured as test sample, feature vector is extracted with the same method, and be input to trained DBN Classifier is realized and is identified to the classification of ancient pottery and porcelain image to be measured and disconnected source, the division of history into periods.
The invention has the following advantages that
The present invention passes through the configuration feature space and the multiple characteristic quantities of glaze colours information space for extracting ancient pottery and porcelain image, then Characteristic quantity is input to SAE-DBN network, Fusion training is carried out to characteristic quantity, point of ancient pottery and porcelain is finally realized according to training result Class and disconnected source, division of history into periods identification provide objective, reliable, quantization, accurate method for the disconnected source division of history into periods identification of ancient pottery and porcelain.
Detailed description of the invention
Fig. 1 is algorithm flow block diagram of the invention；
Fig. 2 is preprocessing module algorithm flow chart
Fig. 3 is configuration characteristic extraction procedure figure；
Fig. 4 is main view correction module schematic diagram；
Fig. 5 is ancient pottery and porcelain color feature extracted procedure chart；
Fig. 6 is the disconnected source division of history into periods module flow diagram of ancient pottery and porcelain.
Fig. 7 is ancient pottery and porcelain multiple features fusion submodule figure.
Specific embodiment
The present invention will be described in detail With reference to embodiment.
The source periodization method as shown in fig.1, a kind of ancient pottery and porcelain based on multicharacteristic information fusion of the present invention is broken, packet The disconnected source division of history into periods module of preprocessing module, characteristic extracting module, ancient pottery and porcelain is included, may be implemented to complete it according to a width antique pottery porcelain photograph Disconnected source division of history into periods detection.
It is clearly ancient to obtain that the preprocessing module uses secondary treatment method to carry out background separation to ancient pottery and porcelain image Ceramic image is simultaneously sent to characteristic extracting module；Characteristic extracting module is used for two feature skies of type, glaze colours in ancient pottery and porcelain image Between in extract two kinds of ancient pottery and porcelain type, glaze colours characteristic quantities simultaneously；The ancient pottery and porcelain is broken source division of history into periods module, is to utilize SAE-DBN net Network model is completed to merge the information of the characteristic quantity of the ancient pottery and porcelain image, realizes that the classification of ancient pottery and porcelain image and ancient pottery and porcelain are disconnected Source, the division of history into periods identify.
1, ancient pottery and porcelain image preprocessing
Referring to fig. 2, image pre-processing module is used to complete the prospect background separation of ancient pottery and porcelain image, and ancient pottery and porcelain is an only nothing Two gem of art, foundation one of of the configuration as the disconnected source division of history into periods, therefore accurate complete extraction are to ancient pottery and porcelain edge wheel Wide information is the prerequisite of subsequent classification and disconnected source division of history into periods identification, but the often ancient pottery and porcelain target object of ancient pottery and porcelain image and back The more difficult differentiation of scape, the boundary region of grey scale change are not it is obvious that the false boundary pixel that generates with noise of background is for mentioning Ancient pottery and porcelain target object is taken to have some impact on, general Boundary extracting algorithm is difficult to get complete clearly ancient pottery and porcelain type Edge image, therefore the preprocessing module completes mention secondary to the effective information of ancient pottery and porcelain image using secondary treatment method It takes, is handled comprising processing for the first time with second；
Processing for the first time is utilization " pyramid " operator to tri- color components of RGB point of each pixel of ancient pottery and porcelain image Not carry out convolution sum processing, then by choose appropriate threshold, extract a first colour edging i.e. edge image；
Specific method is RGB tri- as each pixel in core and ancient pottery and porcelain image by 5*5 pyramid template Color component does convolution sum operation respectively, chooses suitable threshold value then to extract edge.It is directionless due to pyramid operator Property, edge detection effect in all directions is all identical, can detecte accurate edge.The 5*5 pyramid template is as follows:
R, G, B color component of pyramid template and each pixel of ancient pottery and porcelain image carry out field convolution algorithm respectively, The new pixel value of pixel centered on its convolution sum, more by tri- components of RGB treated respectively edge detection effect Accurately, picture in its entirety can obtain first colour edging, referred to as an edge image after pyramid mask convolution operation.
Second of processing is to realize ancient pottery and porcelain display foreground background separation by mask matching method.It calculates first each Edge image is transformed into gray level image by the weighted grey-value of tri- color components of RGB of pixel, and passes through Value-mean filter method contour elimination edge noise jamming, then carry out characteristic information extraction process i.e. secondary edge detected again Journey improves picture contrast and improves the clarity of iconic element, obtains complete ancient pottery and porcelain binaryzation edge graph clearly Composition operation is carried out as being used as ancient pottery and porcelain mask images, then by ancient pottery and porcelain mask images and original image, can be difficult in former background Prospect background separation is realized in complete extraction ancient pottery and porcelain implements target area in isolated ancient pottery and porcelain image.
The intermediate value-mean filter method is the algorithm for combining median filtering method with mean filter method, is effectively overcome The intrinsic defect of two kinds of algorithms itself, not only protects the detail section of ancient pottery and porcelain image, but also eliminate noise well very well Point.Its algorithm are as follows: take out odd number of pixels data from a sampling window of image and press the big minispread of gray scale, select new arrangement sequence The grey scale pixel value of centre three of column simultaneously seeks its average value as output pixel value.The pixel value of the method output is closer to picture To eliminate isolated noise point, output pixel value calculates as follows plain true value: set by one group of pixel by gray value from it is small to Sequence f is lined up greatly1(x,y),f2(x,y),Λ,fi-1(x,y),fi(x,y),fi+1(x,y),Λ,fn(x, y) takes out most intermediate three A pixel point value is respectively fi-1(x,y),fi(x,y),fi+1(x, y), wherein fi(x, y) is the sequence bosom point value newly arranged, Take its average value as output pixel point again.
G (x, y) indicates that gray value of the filtered image at (x, y), ∑ { } are indicated to ancient pottery and porcelain image pixel in formula Most intermediate three pixel gray values summation after wicket pixel gray level sequence where point, m is ancient pottery and porcelain image slices vegetarian refreshments Number.
2, characteristic module is extracted
The characteristic extracting module is used to extract ancient pottery and porcelain simultaneously from two type of ancient pottery and porcelain image, color feature spaces Configuration information and ancient pottery and porcelain glaze colours information, the two characteristic quantities can be as the obvious characteristics of the times of reflection ancient pottery and porcelain and regions The critical index of feature.
(1) ancient ceramics structure
Ancient pottery and porcelain external periphery outline line contains all information of ancient ceramics structure, therefore it is special to extract accurate configuration Sign is most important for the ancient pottery and porcelain source division of history into periods of breaking.The process for extracting contour fitting function is corrected by profile extraction module, main view Module, profile function fitting module three parts image processing operations composition, specific as follows referring to Fig. 3:
A, upper top, foot bounding ellipse function, side wheel are fitted by the ancient pottery and porcelain edge contour information extracted respectively Profile, wherein elliptic function is fitted using least square method, obtains accurate elliptic parameter；
B, ancient pottery and porcelain image main view corrects
Referring to fig. 4, using the elliptic function information of fitting, the shooting angle θ between camera and horizontal direction is sought.
Shooting angle calculation method is as follows:
In formula, θ indicates the angle between camera and horizontal direction, and a indicates the major semiaxis length of the fitted ellipse, b table Show the semi-minor axis length of the fitting bounding ellipse.
There are following relationships with original image for projected image:
The shooting angle asked using formula (1) corrects the ancient pottery and porcelain contour images according to formula (2), and I ' (x, y) is projection Ancient pottery and porcelain image, I (x, y) both corrected rear front view image to obtain ancient pottery and porcelain.
C, image border profile function is fitted
Edge contour Function Fitting is carried out to image is faced, the non-linear of ancient pottery and porcelain implements is approached using BP neural network Edge contour curve chooses optimum network structural parameters, obtains the weight and threshold value between each node by training, obtains meeting essence Desired optimum fit curve is spent, by ancient pottery and porcelain contour curve acquisition multiple type the amount of characteristic parameter of ancient pottery and porcelain: inflection point number n, Area perimeter fractal box fac, the height of neck/stomach heightAncient pottery and porcelain foot area/upper top areaReconstruct is three-dimensional Ancient pottery and porcelain body surface area/ancient pottery and porcelain volumeThese parameters directly reflect ancient ceramics structure, while inflection point information can be with Reflect the concavity and convexity of matched curve, i.e. reflection ancient pottery and porcelain curved shape information；Area perimeter parting box counting dimension reflects ancient pottery and porcelain shape The complexity of shape；The parameter is formed into assemblage characteristic vector to characterize ancient ceramics structure feature.Wherein area perimeter Fractal box facIt calculates as follows:
In formula, s --- the area that ancient pottery and porcelain outline curve is enclosed, l --- ancient pottery and porcelain outline line curve is enclosed Perimeter.facReflect the complexity of ancient pottery and porcelain outline shape.
(2) ancient pottery and porcelain glaze colours information
Referring to Fig. 5, ancient pottery and porcelain glaze color characteristic is extracted using piecemeal color histogram method is improved.By original image rgb color mould Formula is transformed into HSV (hue, saturation, value) color space, and three components of HSV color space are independent from each other, energy It reaches using the vision system of people to the perception degree of color as the model for extracting color characteristic, it is bright from tone, shade, color Image secretly is described, tone and saturation degree component represent color information.Suitable tone is chosen in HSV color space and is satisfied With degree high-low threshold value, accurately to extract image color information；Method is first by ancient pottery and porcelain image segmentation at several pockets To improve accuracy in computation, then ancient pottery and porcelain RGB color is transformed into hsv color space, and by hsv color space H, two color intervals of S are divided into several small color interval i.e. color levels, calculate H, S of each ancient pottery and porcelain image fritter Pixel quantity of two color components in each color level obtains color histogram, and horizontal axis indicates color level range, the longitudinal axis Indicate pixel quantity；
To the color component feature of each segmented image block in whole picture ancient pottery and porcelain image, weighted average is handled respectively again, is obtained The color histogram feature of entire image calculates the color assemblage characteristic vector of its comentropy and energy as ancient pottery and porcelain image, That is ancient pottery and porcelain glaze colours information.
The Information Entropy Features and energy feature of ancient pottery and porcelain color component histogram feature calculate as follows:
1. Information Entropy Features
In formula, E (h), E (s) respectively indicate the comentropy in two channels H, S, and n, m are respectively indicated two colors of H, S point The color level areal of amount, wherein pi=ni/N,piIndicate that pixel falls in the pixel of i-th of color level in H color component The percentage accounted in whole picture ancient pottery and porcelain image, and qj=mj/ M, qjIndicate that pixel falls in j-th of color level in S color component The percentage that is accounted in whole picture ancient pottery and porcelain image of pixel.The comentropy of color histogram is to describe each face of ancient pottery and porcelain image Distribution situation of the form and aspect to whole picture ancient pottery and porcelain image.
2. energy feature
M (h), m (s) respectively represent the energy value of two color component histogram features of H, S in formula.Reflect ancient pottery and porcelain figure As the uniformity coefficient of distribution of color.
3, the disconnected source division of history into periods module of ancient pottery and porcelain
Referring to Fig. 6, the disconnected source division of history into periods module of ancient pottery and porcelain is that the training of DBN classifier, root are completed using SAE-DBN network model The classification and the disconnected source division of history into periods to ancient pottery and porcelain image to be measured are completed according to training result.Detailed process is to make database ancient pottery and porcelain image For training sample, ancient pottery and porcelain image to be detected passes through the above method as test sample, by the ancient pottery and porcelain training sample for having label Configuration feature and color characteristic are extracted, its characteristic quantity is input to depth confidence network and is trained, is completed to described The information of the characteristic quantity of ancient pottery and porcelain image merges, and multicharacteristic information merges submodule referring to Fig. 7, by ancient pottery and porcelain color feature space Information is carried out with multiple characteristic quantities of configuration feature space to merge.Sparse self-encoding encoder (SAE) is recycled to merge information Characteristic quantity carry out dimension-reduction treatment to obtain the description of more better than original multicharacteristic information fused data feature, obtain trained DBN classifier, then ancient pottery and porcelain image to be detected is input to trained DBN classifier and is trained, it is tied according to training Fruit, which can be realized, identifies the classification of ancient pottery and porcelain image to be measured and its disconnected source, the division of history into periods.
The contents of the present invention are not limited to cited by embodiment, and those of ordinary skill in the art are by reading description of the invention And to any equivalent transformation that technical solution of the present invention is taken, all are covered by the claims of the invention.

Claims (7)

  1. The source periodization method 1. the ancient pottery and porcelain based on multicharacteristic information fusion is broken, it is characterised in that:
    Ancient pottery and porcelain implements image is extracted from ancient pottery and porcelain image and realizes background separation, is partitioned into accurate ancient pottery and porcelain image, then right Ancient pottery and porcelain image is filtered denoising and obtains clearly image；
    Type structure feature, glaze color characteristic are extracted simultaneously from two type in image, color feature spaces, it is defeated as characteristic quantity Enter to SAE-DBN network the layer-by-layer extraction for completing information fusion and characteristic quantity to the characteristic quantity of the ancient pottery and porcelain image and takes out As；
    Ancient pottery and porcelain image to be measured is input to trained DBN classifier to be trained, ancient pottery and porcelain is completed by training result Classification and the disconnected source of ancient pottery and porcelain, the division of history into periods identify.
  2. The source periodization method 2. the ancient pottery and porcelain according to claim 1 based on multicharacteristic information fusion is broken, it is characterised in that:
    The method includes preprocessing module, characteristic extracting module, the disconnected source division of history into periods modules of ancient pottery and porcelain；
    It is clearly ancient to obtain that the preprocessing module uses secondary treatment method to carry out prospect background separation to ancient pottery and porcelain image Ceramic image is simultaneously sent to characteristic extracting module；
    Characteristic extracting module be used in two feature spaces of type, glaze colours of ancient pottery and porcelain image simultaneously extract ancient pottery and porcelain type, Two kinds of characteristic quantities of glaze colours；
    The ancient pottery and porcelain is broken source division of history into periods module, is to complete the characteristic quantity to the ancient pottery and porcelain image using SAE-DBN network model Information fusion, by characteristic quantity it is layer-by-layer extraction be abstracted, may finally realize the classification and ancient pottery and porcelain of ancient pottery and porcelain image The disconnected source division of history into periods identifies.
  3. The source periodization method 3. the ancient pottery and porcelain according to claim 2 based on multicharacteristic information fusion is broken, it is characterised in that:
    The preprocessing module use information second extraction method, including processing for the first time and second of processing, for realizing Gu The prospect background of ceramic image is separated to obtain clearly ancient pottery and porcelain target object image；
    The first time processing is utilization " pyramid " operator to tri- color components of RGB point of each pixel of ancient pottery and porcelain image Not carry out convolution sum processing, then choose appropriate threshold extract edge, obtain a first colour edging profile i.e. edge image；
    Second of processing is to realize ancient pottery and porcelain display foreground background separation by mask matching method, calculates each pixel first Edge image is transformed into gray level image by the weighted grey-value of tri- color components of RGB of point, then equal by intermediate value- Value filtering method eliminates the noise jamming in image, carries out characteristic information extraction process, that is, second edge extraction process again, extracts The complete edge binary images of ancient pottery and porcelain are known as ancient pottery and porcelain mask images, then by ancient pottery and porcelain mask images and the compound behaviour of original image Make, can realize prospect in complete extraction ancient pottery and porcelain implements region in the ancient pottery and porcelain image that former background and antique pottery porcelain body are difficult to differentiate between Background separation.
  4. The source periodization method 4. the ancient pottery and porcelain according to claim 2 based on multicharacteristic information fusion is broken, it is characterised in that:
    The extraction ancient ceramics structure information process is fitted by profile extraction module, main view correction module, profile function Module three parts image processing operations composition；
    The profile extraction module is the ancient pottery and porcelain image denoising and binary conversion treatment after separating to the extraction prospect background, then Extract top on edge contour, including ancient pottery and porcelain implements, foot ellipse, outline line；
    The ancient pottery and porcelain image border profile extracted according to the main view correction module method determines ancient pottery and porcelain shooting angle, Front view, i.e. ancient pottery and porcelain front view are restored to ancient pottery and porcelain image rectification；
    The profile function fitting module using the BP neural network functional characteristics approached of height to ancient pottery and porcelain front view again into Row edge contour extracts and Function Fitting；The neural network needs to choose optimum network structural parameters, obtains by training Weight and threshold value between each node, obtain the optimum fit curve for meeting required precision, obtain antique pottery by ancient pottery and porcelain contour curve Multiple type the amount of characteristic parameter of porcelain: inflection point number n, area perimeter fractal box fac, the height of neck/stomach heightIt is ancient Ceramic foot area/upper top areaReconstruct three-dimensional ancient pottery and porcelain body surface area/ancient pottery and porcelain volumeGenerated one group of combination Feature vector is to characterize ancient ceramics structure feature.
  5. The source periodization method 5. the ancient pottery and porcelain according to claim 2 based on multicharacteristic information fusion is broken, it is characterised in that:
    The extraction of the ancient pottery and porcelain glaze colours information, for ancient pottery and porcelain image segmentation is first improved calculating at several pockets Accuracy, then ancient pottery and porcelain RGB color is transformed into hsv color space, and by two colors of H, the S in hsv color space Section is divided into several small color interval i.e. color levels, calculates two colors of H, S point of each ancient pottery and porcelain image fritter It measures the pixel quantity in each color level and obtains color histogram, horizontal axis indicates color level range, and the longitudinal axis indicates pixel number Amount；To the color component feature of each segmented image block in whole picture ancient pottery and porcelain image, weighted average is handled respectively again, obtains whole picture The color histogram feature of image calculates the color assemblage characteristic vector of its comentropy and energy as ancient pottery and porcelain image, i.e., ancient Ceramic glaze colours information.
  6. The source periodization method 6. the ancient pottery and porcelain according to claim 2 based on multicharacteristic information fusion is broken, it is characterised in that:
    The ancient pottery and porcelain is broken source division of history into periods module, is broken source division of history into periods detection using ancient pottery and porcelain is carried out based on SAE-DBN network model；From Gu Ceramic type feature space and color feature space extract ancient ceramics structure feature and color characteristic information simultaneously respectively, will It generates assemblage characteristic vector, breaks the judgment basis of the source division of history into periods as ancient pottery and porcelain；The characteristic quantity extracted is input to SAE-DBN network, completes to merge the information of the characteristic quantity of the ancient pottery and porcelain image, is merged using sparse self-encoding encoder to information Characteristic quantity carry out dimension-reduction treatment to obtain the description of better feature, then characteristic is input to training in depth confidence network DBN classifier completes the classification of ancient pottery and porcelain image to be measured according to training result and disconnected source, the division of history into periods identifies.
  7. The source periodization method 7. the ancient pottery and porcelain according to claim 6 based on multicharacteristic information fusion is broken, it is characterised in that:
    The concrete methods of realizing of the disconnected source division of history into periods of ancient pottery and porcelain is to extract using the ancient pottery and porcelain image in database as training sample The feature vector of training sample is input to SAE-DBN network and completes network training and generate trained DBN classifier, then will be to Antique pottery porcelain photograph is surveyed as test sample, feature vector is extracted with the same method, and is input to trained DBN classification Device is realized and is identified to the classification of ancient pottery and porcelain image to be measured and disconnected source, the division of history into periods.
CN201811451703.0A 2018-11-30 2018-11-30 Ancient ceramic source breaking and generation breaking method based on multi-feature information fusion Active CN109583376B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811451703.0A CN109583376B (en) 2018-11-30 2018-11-30 Ancient ceramic source breaking and generation breaking method based on multi-feature information fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811451703.0A CN109583376B (en) 2018-11-30 2018-11-30 Ancient ceramic source breaking and generation breaking method based on multi-feature information fusion

Publications (2)

Publication Number Publication Date
CN109583376A true CN109583376A (en) 2019-04-05
CN109583376B CN109583376B (en) 2022-09-27

Family

ID=65925562

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811451703.0A Active CN109583376B (en) 2018-11-30 2018-11-30 Ancient ceramic source breaking and generation breaking method based on multi-feature information fusion

Country Status (1)

Country Link
CN (1) CN109583376B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363752A (en) * 2019-07-08 2019-10-22 创新奇智(青岛)科技有限公司 A kind of ready-made clothes material defects simulation generation method, computer-readable medium and system
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
CN111339974A (en) * 2020-03-03 2020-06-26 景德镇陶瓷大学 Method for identifying modern ceramics and ancient ceramics
CN115340408A (en) * 2022-09-05 2022-11-15 宁波职业技术学院 Method for repairing ancient porcelain ware without damage during punching
CN115587298A (en) * 2021-07-05 2023-01-10 中国矿业大学(北京) Historical Jingdezhen blue and white porcelain age discrimination method based on deep learning

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018084507A1 (en) * 2016-11-02 2018-05-11 김명진 Method and system for authenticating ancient earthenware and china using absolute dating technique
CN108320282A (en) * 2018-01-25 2018-07-24 陕西科技大学 The ceramic wall and floor bricks surface defect detection apparatus and its method of multicharacteristic information fusion

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018084507A1 (en) * 2016-11-02 2018-05-11 김명진 Method and system for authenticating ancient earthenware and china using absolute dating technique
CN108320282A (en) * 2018-01-25 2018-07-24 陕西科技大学 The ceramic wall and floor bricks surface defect detection apparatus and its method of multicharacteristic information fusion

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
熊露等: "基于特征提取的古陶瓷鉴定研究", 《中国陶瓷》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363752A (en) * 2019-07-08 2019-10-22 创新奇智(青岛)科技有限公司 A kind of ready-made clothes material defects simulation generation method, computer-readable medium and system
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
CN111339974A (en) * 2020-03-03 2020-06-26 景德镇陶瓷大学 Method for identifying modern ceramics and ancient ceramics
CN111339974B (en) * 2020-03-03 2023-04-07 景德镇陶瓷大学 Method for identifying modern ceramics and ancient ceramics
CN115587298A (en) * 2021-07-05 2023-01-10 中国矿业大学(北京) Historical Jingdezhen blue and white porcelain age discrimination method based on deep learning
CN115340408A (en) * 2022-09-05 2022-11-15 宁波职业技术学院 Method for repairing ancient porcelain ware without damage during punching

Also Published As

Publication number Publication date
CN109583376B (en) 2022-09-27

Similar Documents

Publication Publication Date Title
CN109583376A (en) The disconnected source periodization method of ancient pottery and porcelain based on multicharacteristic information fusion
CN107578418B (en) Indoor scene contour detection method fusing color and depth information
CN103927741B (en) SAR image synthesis method for enhancing target characteristics
CN105184787B (en) A kind of license camera and method for portrait scratch figure automatically
CN106651872A (en) Prewitt operator-based pavement crack recognition method and system
CN108090960A (en) A kind of Object reconstruction method based on geometrical constraint
CN108230338B (en) Stereo image segmentation method based on convolutional neural network
CN102902956B (en) A kind of ground visible cloud image identifying processing method
CN102930334B (en) Video recognition counter for body silhouette
CN109815833A (en) A kind of tea point recognition methods based on CCD Yu the feature extraction of three-dimensional laser sensor information fusion
CN110738676A (en) GrabCT automatic segmentation algorithm combined with RGBD data
CN107403168A (en) A kind of facial-recognition security systems
CN106446851A (en) Visible light based human face optimal selection method and system
CN109584281B (en) Overlapping particle layering counting method based on color image and depth image
CN103035013A (en) Accurate moving shadow detection method based on multi-feature fusion
CN111354077B (en) Binocular vision-based three-dimensional face reconstruction method
CN104361574B (en) No-reference color image quality assessment method on basis of sparse representation
CN105205804A (en) Caryon-cytolymph separation method and apparatus of white blood cells in blood cell image, and classification method and apparatus of white blood cells in blood cell image
CN103870808A (en) Finger vein identification method
CN109086723A (en) A kind of method, apparatus and equipment of the Face datection based on transfer learning
CN105701805B (en) A kind of pork intramuscular fat content lossless detection method based on computer vision
CN110544300B (en) Method for automatically generating three-dimensional model based on two-dimensional hand-drawn image characteristics
CN109559324A (en) A kind of objective contour detection method in linear array images
CN109214298A (en) A kind of Asia women face value Rating Model method based on depth convolutional network
CN104143077B (en) Pedestrian target search method and system based on image

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
GR01 Patent grant
GR01 Patent grant