CN105844583A - Portrait stone crack intelligence extraction and virtual restoration method - Google Patents

Abstract

The invention discloses a portrait stone crack intelligence extraction and virtual restoration method. The method comprises the following steps of 1) acquiring a digital image of a portrait stone and preprocessing the digital image of the portrait stone; 2) for the preprocessed digital image of the portrait stone, using a crack extraction algorithm based on an editing distance to carry out crack disease calibration so as to acquire a mask image, and then acquiring a labeled graph of a crack disease in the portrait stone according to the mask image; 3) using a Criminisi restoration algorithm based on a MRF matching criterion to carry out digital restoration on the digital image of the portrait stone according to the labeled graph of the crack disease in the portrait stone which is acquired from the step 2), generating the restored digital image of the portrait stone and then displaying the restored digital image of the portrait stone. The method can be used to satisfy the restoration of the digital image of the portrait stone.

Description

A kind of stone relief crack intelligent extraction and virtual restorative procedure

Technical field

The invention belongs to digital image processing field, relate to a kind of stone relief crack intelligent extraction and void Intend restorative procedure.

Background technology

Stone relief, belongs to the one of ancient wall, mainly Han dynasty underground burial chamber, graveyard ancestral hall, The building structure stone of the architectural carving picture pictures such as tomb fault and mausoleum fault.They are not only research and development based on inheritance The carrier of China's tradition art of stone engraving, and be the treasure of the researchs such as history, politics, military affairs, art Your data.Different from paper document, stone relief document has only existing copy, and once damage can not be again Raw, however currently inadequate to the attention degree of stone relief document, add natural ring around stone relief The complexity in border, immature, the restriction of the factor such as fund plaque is weary of sci-tech protection method, Bu Shaohua Image-stone document is all subject to damage in various degree, it is difficult to meet experts and scholars and it is done follow-up study Needing, the protection of stone relief becomes the emphasis of domestic scholars research with repairing.

Because buring environment and the impact of later stage anthropic factor, stone relief can produce disease, mainly wrap Include fracture, crack, wind-force crack, cracking, incompleteness and cement repairing etc..As time goes on, The infringement of stone relief will be more serious.The poor efficiency of traditional protection means and the shortage of protection personnel, Constrain the development of China's stone relief protection cause.Digitized image have can store, can transmit, Can process, renewable, available, the feature such as can share, the just right stone relief that compensate for The defect such as non-renewable, irreplaceable and fragile.Image processing techniques is utilized to assist stone relief Protection is possible not only to reduce the difficulty of stone relief protection, but also reduces anthropic factor to stone relief Damage, be digitized stone relief preserving, and utilize computer technology that it is carried out virtual reparation, To guide actual repair and exhibition, be Conservative restoration stone relief with construction historical relic digital archives is rigid Demand.

At present, although have an algorithm of some fracture region labelings, but these algorithms generally require with Manually combine and demarcate the most accurate.The most accurately, position automatically and extract stone relief crack district Territory is still the emphasis of scholar's research.

Although there is now the method much repaired about digital picture, but owing to image repair technology is Infer the unknown message lacked completely according to Given information, therefore repair result and original image itself Feature have the biggest association, so the most also not having which kind of algorithm to be suitable for all of breakage image.State Inside and outside less to the research of stone relief image repair, a kind of algorithm disclosure satisfy that portrait The reparation of stone complex texture.

Summary of the invention

It is an object of the invention to the shortcoming overcoming above-mentioned prior art, it is provided that a kind of portrait stony fracture Seam intelligent extraction and virtual restorative procedure, the method meets the reparation of stone relief digital picture.

For reaching above-mentioned purpose, stone relief crack intelligent extraction of the present invention and virtual reparation side Method comprises the following steps:

1) obtain the digital picture of stone relief, then the digital picture of stone relief is carried out pretreatment；

2) digital picture to pretreated stone relief utilizes crack extract based on editing distance to calculate Method carries out crack disease demarcation, obtains mask image, then obtains in stone relief according to mask image and split The labelling figure of seam disease；

3) Criminisi based on MRF matching criterior is utilized to repair algorithm to step 2) obtain In stone relief, the labelling figure of crack disease carries out numeral reparation, and generates the numeral of the stone relief after reparation Image, then shows the stone relief digital picture after repairing.

Step 1) in the digital picture of stone relief carried out the process of pretreatment be: calculate stone relief Digital picture in the weight of each pixel, then the weight of each pixel is clicked on corresponding pixel Row point multiplication operation, then the result of the erosion algorithm process points multiplication in recycling morphology, convex Show the crack disease region in the digital picture of stone relief.

Criminisi based on MRF matching criterior is utilized to repair algorithm to step 2) picture that obtains In image-stone the labelling figure of crack disease carry out numeral reparation concrete operations be:

1a) according to each picture during the labelling figure of crack disease calculates crack disease edges of regions in stone relief The priority of element block, the highest block of pixels of selecting priority is as block of pixels to be repaired；

2a) obtain the Bayes conditional probability that block of pixels to be repaired meets with its neighborhood system, i.e.

$P(X=x|Y=y)=\frac{P(Y=y|X=x)P(X=x)}{P(Y=y)}---\left(5\right)$

Wherein, X is block of pixels to be repaired, and Y is the neighborhood of known image and block of pixels X to be repaired System, about the posterior probability of block of pixels X to be repaired when P (X=x | Y=y) is known Y, P (X=x) is the prior probability of image, the best estimate of block of pixels the most to be repairedFor:

$\hat{x}=Arg\underset{x}{max}\left(\ln p\right(Y=y|X=x)+\ln p(X=x\left)\right)---\left(6\right)$

Optimum 3 × 3 match block of search in the digital picture of stone relief, wherein, optimum 3 × 3 match block The best estimate of average gray value and block of pixels to be repairedBetween difference minimum；

3a) optimum 3 × 3 match block is copied in block of pixels to be repaired, complete block of pixels to be repaired Reparation；

4a) select next block of pixels as block of pixels to be repaired according to the size of priority, lay equal stress on Multiple step 2a) and 3a)；

5a) repeat step 4a), until each block of pixels in all slits disease region has all been repaired Till.

During known Y, posterior probability p (Y=y | X=x) about block of pixels X to be repaired is:

$p(Y=y|X=x)={\left(\frac{1}{2{\mathrm{\π\δ}}^2}\right)}^{N_1{N}_2/2}\exp \{-\frac{1}{2{\mathrm{\δ}}^2}\underset{m=1}{\overset{M}{\Σ}}{(y_s-q_m)}^2\}---\left(7\right)$

Wherein, δ²For gradation of image variance, image is made up of M region, q_mFor each region Gray average, N₁And N₂Length and width for image.

The expression formula of the prior probability p (X=x) of image is:

$p(X=x)=\frac{1}{Z}{e}^{-U\left(X\right)/T}---\left(8\right)$

Wherein, T is thermal constant, and Z is normaliztion constant, and U (X) is the global energy letter of image Number, and

$U\left(X\right)=\underset{p\∈v}{\Σ}{V}_p\left(x_p\right)+\underset{(p,q)\∈\mathrm{\ϵ}}{\Σ}{V}_{pq}(x_p,x_q)---\left(9\right)$

Wherein, V_p(x_p) it is to the likelihood energy of pixel one label of distribution, V_pq(x_p,x_q) it is The prior model energy of two neighbor two labels of distribution, (p, q) is neighborhood territory pixel pair, and ε is 4 neighborhood territory pixels in system are to set.

Step 2) also include storing the labelling figure of crack disease in stone relief.

Step 3) also include storing the stone relief digital picture after repairing.

The method have the advantages that

Stone relief crack intelligent extraction of the present invention and virtual restorative procedure in repair process, Crack extract algorithm based on editing distance is utilized to carry out crack disease demarcation, it is achieved to portrait work of art created with stones Crack disease in face is automatically positioned and demarcates, and solves the most artificial stone relief crack of demarcating and is forbidden True and inefficient problem, after tested, the stated accuracy of the present invention reaches more than 95%.It addition, The present invention utilizes Criminisi based on MRF matching criterior to repair the algorithm digitized map to stone relief As carrying out digital virtual reparation, the effect of reparation is preferable, when solving reparation complex texture stone relief The problem that blocking effect is serious, precision is relatively low.The present invention can show picture by empty paraprotehetic mode The repair process of image-stone and structure, provide the most perfect reference for actual repair work, it is to avoid repair The mixed and disorderly disorder phenomenon occurred during Fu.

Further, the present invention is when repairing each block of pixels, according to the priority of each block of pixels The height of grade is repaired successively, during by repairing each block of pixels, chooses optimum 3 × 3 Join block, make the best estimate of 3 × 3 match block block of pixels to be repairedBetween difference minimum, then By optimum 3 × 3 match block, block of pixels is replaced again, repairs precision height, meet stone relief pair Repair the requirement of precision.

Accompanying drawing explanation

Fig. 1 is stone relief crack disease artwork used in the present invention；

Fig. 2 is to calculate the design sketch of stone relief after weight in the present invention；

The design sketch of point multiplication operation result in Fig. 3 present invention；

Fig. 4 is the design sketch in the present invention after erosion algorithm processes；

Fig. 5 is the labelling figure of crack disease in stone relief in the present invention；

Fig. 6 is the design sketch in the present invention after stone relief reparation.

Detailed description of the invention

Below in conjunction with the accompanying drawings the present invention is described in further detail:

Stone relief crack intelligent extraction of the present invention and virtual restorative procedure comprise the following steps:

1) obtain the digital picture of stone relief, then the digital picture of stone relief is carried out pretreatment；

2) pretreated stone relief digital picture is utilized crack extract algorithm based on editing distance Carry out crack disease demarcation, obtain mask image, then obtain crack in stone relief according to mask image The labelling figure of disease；

3) Criminisi based on MRF matching criterior is utilized to repair algorithm to step 2) obtain In stone relief, the labelling figure of crack disease carries out numeral reparation, and generates the numeral of the stone relief after reparation Image, then shows the stone relief digital picture after repairing.

Step 1) in the digital picture of stone relief carried out the process of pretreatment be: calculate stone relief Digital picture in the weight of each pixel, then the weight of each pixel is clicked on corresponding pixel Row point multiplication operation, then the result of the erosion algorithm process points multiplication in recycling morphology, convex Show the crack disease region in the digital picture of stone relief, wherein, to original image I_oIn each Pixel carries out the calculating of weight, obtains weight map as I_w, shown in specific formula for calculation such as formula (1):

I_w(x, y)=exp (-I_o(x,y)) (1)

Then by original image I_oWith weight map as I_wCarry out point multiplication operation, in the new images generated Crack area can be marked as the region that gray value is relatively low, but due to light etc. around crack area The impact of factor, gray value also can be smaller, therefore the erosion operation in recycling morphology afterwards Image carries out single treatment again, and in erosion operation, structural element selects the length of side to be the rectangle of 16, To accurate crack disease region.

Step 2) concrete operations be:

First digital picture through the stone relief of pretreatment before is divided into the region of multiple 3 × 3, Being the vector of 1 × 9 by each region 3 × 3 matrix conversion, zoning φ vector sum is adjacent on the right side of it Region φ_rVectorial and that lower section is adjacent region φ_bThe editing distance of vector, editing distance calculates such as Under:

Calculate two vectorial φ_mAnd φ_rnEditing distance, wherein, m and n be respectively two vectorial Length, structural matrix d [m+2, n+2], this matrix d [m+2, n+2] is m+2 row, the square of n+2 row Battle array.

If m=n=3, then matrix d [m+2, n+2] is as shown in table 1:

Table 1

From (3,3), lattice start, start calculate, value have following three kinds may:

Work as φ_r1Equal to φ₁, then it is upper left numeral, is otherwise upper left numeral+1；Left number Word+1；Upper values+1；

Three above value is compared, takes minima and be assigned to d [3,3], cycle calculations matrix Each value in d [m+2, n+2], then in table 1 the value i.e. d [5,5] in the lower right corner be just two vectorial Editing distance.

By calculated vector φ and φ_r, φ and φ_bBefore the meansigma methods assignment of two editing distances Region φ, obtains the edge graph of crack area；Calculate the strong limit obtaining in image of edge strength again Edge, then the region completion in edge is obtained complete crack by the closed operation in recycling morphology Region.

Step 3) in utilize Criminisi based on MRF matching criterior repair algorithm according to step 2) In the stone relief obtained, the labelling figure of crack disease carries out numeral reparation to the digital picture of stone relief Concrete operations are:

1a) according to each picture during the labelling figure of crack disease calculates crack disease edges of regions in stone relief The priority of element block, the highest block of pixels of selecting priority is as block of pixels to be repaired；

Wherein, the calculating of priority is carried out according to formula (2)

Priority: P (p)=C (p) × D (p) (2)

Confidence level:

In formula (3), on border, area to be repaired, confidence level C (p) of pixel p is area to be repaired Ψ_p Known pixels sum with its contained by the ratio of sum of all pixels, Size (Ψ_p) it is Ψ_pContained sum of all pixels.

Data item:

Weighing data item D (p) of edge strength at pixel p in formula (4) is limit, area to be repaired The unit normal vector n at p is put in boundary_pWith isophote vectorProduct, wherein, α is normalizing Changing parameter, generally for gray level image α=255, ε is constant, it is to avoid D (p) is 0.

2a) obtain block of pixels to be repaired and meet Bayes conditional probability with its neighborhood system, i.e.

$P(X=x|Y=y)=\frac{P(Y=y|X=x)P(X=x)}{P(Y=y)}---\left(5\right)$

Wherein, X is block of pixels to be repaired, and Y is the neighborhood of known image and block of pixels X to be repaired System, about the posterior probability of block of pixels X to be repaired when P (X=x | Y=y) is known Y, P (X=x) is the prior probability of image, the best estimate of block of pixels the most to be repairedFor:

$\hat{x}=Arg\underset{x}{max}\left(\ln p\right(Y=y|X=x)+\ln p(X=x\left)\right)---\left(6\right)$

Optimum 3 × 3 match block of search in the digital picture of stone relief, wherein, optimum 3 × 3 match block The best estimate of average gray value and block of pixels to be repairedBetween difference minimum；

3a) optimum 3 × 3 match block is copied in block of pixels to be repaired, complete block of pixels to be repaired Reparation；

4a) select next block of pixels as block of pixels to be repaired according to the size of priority, lay equal stress on Multiple step 2a) and 3a)；

5a) repeat step 4a), until each block of pixels is all repaiied in all slits disease edges of regions Till completing again.

During known Y, posterior probability p (Y=y | X=x) about block of pixels X to be repaired is:

$p(Y=y|X=x)={\left(\frac{1}{2{\mathrm{\π\δ}}^2}\right)}^{N_1{N}_2/2}\exp \{-\frac{1}{2{\mathrm{\δ}}^2}\underset{m=1}{\overset{M}{\Σ}}{(y_s-q_m)}^2\}---\left(7\right)$

Wherein, δ²For gradation of image variance, image is made up of M region, q_mFor each region Gray average, N₁And N₂Length and width for image.

The expression formula of the prior probability p (X=x) of image is:

$p(X=x)=\frac{1}{Z}{e}^{-U\left(X\right)/T}---\left(8\right)$

Wherein, T is thermal constant, and Z is normaliztion constant, and U (X) is the global energy letter of image Number, and

$U\left(X\right)=\underset{p\∈v}{\Σ}{V}_p\left(x_p\right)+\underset{(p,q)\∈\mathrm{\ϵ}}{\Σ}{V}_{pq}(x_p,x_q)---\left(9\right)$

Wherein, V_p(x_p) it is to the likelihood energy of pixel one label of distribution, V_pq(x_p,x_q) it is The prior model energy of two neighbor two labels of distribution, (p, q) is neighborhood territory pixel pair, and ε is 4 neighborhood territory pixels in system are to set.

It addition, step 2) also include storing the labelling figure of crack disease in stone relief；Step 3) also The digital picture of the stone relief after repairing including storage.

Claims (7)

1. a stone relief crack intelligent extraction and virtual restorative procedure, it is characterised in that include with Lower step:

1) obtain the digital picture of stone relief, then the digital picture of stone relief is carried out pretreatment；

2) pretreated stone relief digital picture is utilized crack extract algorithm based on editing distance Carry out crack disease demarcation, obtain mask image, then obtain crack in stone relief according to mask image sick The labelling figure of evil；

3) Criminisi based on markov random file matching criterior is utilized to repair algorithm to step 2) In the stone relief obtained, the labelling figure of crack disease carries out numeral reparation, and generates the stone relief after reparation Digital picture, then shows the stone relief digital picture after repairing.

Stone relief crack intelligent extraction the most according to claim 1 and virtual restorative procedure, its Be characterised by, step 1) in the digital picture of stone relief carried out the process of pretreatment be: calculate picture The weight of each pixel in the digital picture of image-stone, then by the weight of each pixel and corresponding pixel Carrying out point multiplication operation, then the result of the erosion algorithm process points multiplication in recycling morphology, convex Show the crack disease region in the digital picture of stone relief.

Stone relief crack intelligent extraction the most according to claim 1 and virtual restorative procedure, its It is characterised by, utilizes Criminisi based on MRF matching criterior to repair algorithm to step 2) obtain Stone relief in the labelling figure of crack disease carry out the concrete operations of numeral reparation and be:

1a) according to each pixel during the labelling figure of crack disease calculates crack disease edges of regions in stone relief The priority of block, the highest block of pixels of selecting priority is as block of pixels to be repaired；

2a) obtain the Bayes conditional probability that block of pixels to be repaired meets with its neighborhood system, i.e.

$P(X=x|Y=y)=\frac{P(Y=y|X=x)P(X=x)}{P(Y=y)}---\left(5\right)$

Wherein, X is block of pixels to be repaired, and Y is the neighborhood system of known image and block of pixels X to be repaired System, about the posterior probability of block of pixels X to be repaired when P (X=x | Y=y) is known Y, P (X=x) is The prior probability of image, the best estimate of block of pixels the most to be repairedFor:

$\hat{x}=Arg\underset{x}{max}\left(lnp\right(Y=y|X=x)+lnp(X=x\left)\right)---\left(6\right)$

Optimum 3 × 3 match block of search in the digital picture of stone relief, wherein, optimum 3 × 3 match block The best estimate of average gray value and block of pixels to be repairedBetween difference minimum；

3a) optimum 3 × 3 match block is copied in block of pixels to be repaired, complete block of pixels to be repaired Repair；

4a) select next block of pixels as block of pixels to be repaired according to the size of priority, and repeat Step 2a) and 3a)；

5a) repeat step 4a), until each block of pixels in all slits disease region all repaired into Only.

Stone relief crack intelligent extraction the most according to claim 3 and virtual restorative procedure, its It is characterised by, it is known that during Y, posterior probability p (Y=y | X=x) about block of pixels X to be repaired is:

$p(Y=y|X=x)={\left(\frac{1}{2{\mathrm{\π\δ}}^2}\right)}^{N_1{N}_2/2}\exp \{-\frac{1}{2{\mathrm{\δ}}^2}\underset{m=1}{\overset{M}{\Σ}}{(y_s-q_m)}^2\}---\left(7\right)$

Wherein, δ²For gradation of image variance, image is made up of M region, q_mFor each region Gray average, N₁And N₂Length and width for image.

Stone relief crack intelligent extraction the most according to claim 4 and virtual restorative procedure, its Being characterised by, the expression formula of the prior probability p (X=x) of image is:

$p(X=x)=\frac{1}{Z}{e}^{-U\left(X\right)/T}---\left(8\right)$

Wherein, T is thermal constant, and Z is normaliztion constant, and U (X) is the global energy function of image, And

$U\left(X\right)=\underset{p\∈v}{\Σ}{V}_p\left(x_p\right)+\underset{(p,q)\∈\mathrm{\ϵ}}{\Σ}{V}_{pq}(x_p,x_q)---\left(9\right)$

Wherein, V_p(x_p) it is to the likelihood energy of pixel one label of distribution, V_pq(x_p,x_q) it is two The prior model energy of individual neighbor two labels of distribution, (p, q) is neighborhood territory pixel pair, and ε is system In 4 neighborhood territory pixels to set.

Stone relief crack intelligent extraction the most according to claim 1 and virtual restorative procedure, its It is characterised by, step 2) also include storing the labelling figure of crack disease in stone relief.

Stone relief crack intelligent extraction the most according to claim 1 and virtual restorative procedure, its It is characterised by, step 3) also include storing the stone relief digital picture after repairing.