Image processing methods for the restoration of digitized paintings, A Gupta, V Khandelwal, A Gupta

Tags: Median Filter, MAO, N. Otsu, pp, Nonlinear Digital Filters, A. N. VenetsanoPoulos, L. Vincent, IEEE Int, window size, pixels, ancient paintings, modified version, J. Sc, pixel count, Digital Restoration, local region, Figs, Restorationof DigitizedPaintings Abhilekh Gupta, Abhinav Gupta, thresholding operation, automatic thresholding, binary image, Vineet Khandelwal, colour paintings, BHT, Electronics& Communications, digital paintings, Canny edge detection
Content: INTRODUCTION TECHNIQUESch..Vol. 13, No.3, July-Septembe2r008
ImageProcessing Methodsfor the Restorationof DigitizedPaintings
Abhilekh Gupta, Vineet Khandelwal, Abhinav Gupta and M. C. Srivastava Dept.of Electronics& Communications, JaypeeInstituteof InformationTechnologyUniversity Noida (U.P.),lndia [email protected][,vineet.khandelwala,bhinav.guptam, c.srivastava]@jiit.ac.in
Abstract Severalmethodshave beenproposedfor detectionand removal of cracksin digitized paintings !-51. Cfacks not only deterioratethe quality of painting but also questionits authenticity.In this paper, a morphologicalmethodology(MAO) is proposedwhich is a variant of recently published morphologicalmethodsto identify cracks []. After detectingcracks,a modified adaptivemedian filter (MAMF) is usedto fill the cracks.The orderof the medianfilter to be appliedon crackpixels is computedon the basis of the number of crack pixels in its neighbourhood.This methodologyof detectionand eliminationof cracksin digitizedpaintingsis shownto be very effectivein preserving the edgesalso.
Keywords: Digital image processing,digitized paintings,crack detection,crack filling, modified adaptivemedianfilter, virtual restorationof paintings.
1. Introduction Image processingtechniqueshave recently been applied to analysis, preservation and restoration of artwork. Ancient paintings are cultural heritage for ones country which can be preserved by computer aided analysis and processing. These paintings get deteriorated mainly by an undesired pattern that causes breaks in the paint, or varnish. Such a Pattern Can be rectangular, circular, spider-web, unidirectional,treebranchesandrandom[2] and are usually called cracks. Cracks are caused mainly by aging, drying and mechanicalfactors like vibration,andhumanhandling. Computer aided tools can be used to implement image processingtechniquefor the elimination and detectionof cracks in ancient digital paintings. Though this methodology produces a digitally restored version of the artwork, it can still prove to be a useful guide for historians,and museum curators.A technique that is able to track and fiIl a crack is proposed in [3] but it requiresthe user to manually start
with the initial point of the crackpatternto fill them.A methodfor the eliminationof the cracks using an infraredreflectogramofthe painting is presentedin [a]. In this approach a viscous morphological reconstructiontechnique,based on a-priori information about the thickness of the cracks and its preferred orientation, is assumed for crack elimination. Abas anc Martinez [2] have proposed a technique for the detection and classification of cracks using content based analysis. This method uses a morphological top-hat operator to detect the crack and fuzzy k-meansclustering techniqueto classify the various crack patterns. A similar problem of detectionand filling of crackshas been treatedby Giakoumis and Pitas [5]. Their process first detects the crack using a morphological top-hat operator and then fills them using a trimmed median filter [6], and an anisotropicdiffusion fi lter. Morphological Area Opening (MAO) is one of many techniquesfor obtaining the crack map [7]. In section2 of this paperwe proposea
66
ThammasaItnt. J. Sc. Tech..Vol. 13. No.3. Julv-Sentembe2r008
new crack detection model which utilizes bottom hat operationfollowed by thresholding before applying to MAO for obtaining crack map. Next we presentin section3 a technique for successive filling of the cracks using Modified Adaptive Median Filter (MAMF) on digitized paintings. Concluding remarks are discussedin section4. 2. A New Crack Detection Model Cracks haring low luminancepixels with elongated structural characteristics are consideredas local minima [5]. The proposed crack detectionmodel employing a bottom-hat transform followed by thresholdingand MAO is shownin Fig.1. The detectionprocesshrst involves bottom-hat transform(BHT) over luminancecomponentof the imageto detectdark pixels,given by:
B H T = c / " r n . r n n (-xc)r ( x )
(1)
where cr.1o.1nr(xr)epresents closing of luminancecomponentof the image cr(x) using a structurinselement.
binary image. Instead of using a global
threshold, the crack image can be locally
processed using grid-based automatic
thresholding [2].Thereafter,the MAO process producesa crack map consistingof true cracks.
Thus, the MAO filter removesfrom the binary
image,the componentswith areasmallerthan a
predefinedparametera. This filter is definedas:
'f' " ( a l B c :
B
.
v BB
r
.
.
"f u
"
B
B B , , . o : { x - e , X i , B " _ r u r r " r , " oA, r e a ( x ) 2 a }
where/is the binary image from which an areasmallerthan parametera is removed,using connectivity given by B.. This operator is generalizedto binary images by applying the operatorsuccessivelyon slicesof / takenfrom higherthresholdlevelsto lower thresholdlevels, producing a crack map. The parameter a is selectivelychosenfrom the thresholdedoutput image and B. is taken as S-connected.Three crack maps processedthrough our model are shown in Figs. 2 (b), 3(b), 4(b) and 5(b) along with crackedpainting Figs.2 (a), 3(a), 4(a) and 5(a) (religiousicons from the Byzantineera and otherpaintings).
MorphologicalArea Opening Figure 1. CrackDetectionModel
The BHT operatoris optimized using the following parameters: . The type and size of the structuring element: A square type structuring elementis usedwith a sizeof3 x 3. . Numberof operationisn (l): 1 Theseparametersare carefully chosensuch that misidentificationof cracks does not take place.However,they areof low significancedue to the thresholdingand MAO procedure,which identifiesthe crackmap. A thresholding operation is performed on the output of the BHT image to separate crack pixels from the background. A global thresholdingtechnique[8], operatingdirectly on the BHT histogram, is used to produce the
Figure 2 (a). CrackedPainting(b).CrackMap
61
ThammasaItnt. J. Sc. Tech..Vol. 13, No.3, Julv-Septembe2r008
Figure 3 (a). CrackedPainting(b). CrackMap
Figure 4(a). CrackedPainting(b). CrackMap
Figure 5(a). CrackedPainting(b). Crack Map 3. Modified Adaptive Median Filter (MAMF) for Crack Filling Oncea crackmap is obtainedour next step is to fill the cracks based on the local information from the region surrounding the crackpixel. We proposea new filter known as a Modified Adaptive Median Filter (MAMF) which works on each RGB channel independentlyonly on the crack pixel locations, so that quality of the contentin other pixels is not affected.This nonlinearfilter, in additionto crack filling, preserves the edges of the paintings. Since we intend to use ancient paintings whose original version is not known for detection and elimination of cracks, our methodof filling canbejudged qualitativelyjust by Visual Inspectiononly. The standardmedian filter could be used for filling the crack, but the problem with this method lies on its fixed window size and that there could always be a possibility that crack pixel count in the local region may exceedthe non crackpixel count.This may however,result in replacinga crackpixel by anothercrackpixel, thus, failing in our aim. We thereforeproposea modified version of an adaptivemedian filter
68
ThammasaItnt. J. Sc. Tech..Vol. 13. No.3. Julv-Septembe2r008
with varying the window size surroundingthe crackpixel. This variationdependson the nature of pixels surroundingthe crack pixels in the local region of window. MAMF runs only over the crack pixels so that information in other pixel is kept intact.The sizeof the filter window surroundingeachcrack pixel is evaluatedbased on the numberof crackpixels in the local region of the window. If the numberof crackpixels in the local region exceedssome thresholdvalue, the size of the window is expandedtill it falls below the threshold.In our case,a threshold value is set to be equivalent to 25oh of the number of pixels in the window, having examined the effect of the threshold on the window size. When the number crack pixels in the local region falls below this thresholdlevel, the size of the window satisfyingthe condition is treat-ed as the order /y' of the filter for processingthe crack pixel under observation.1y' will be different for all crack pixels in the painting and is evaluated adaptively. 'Processing'hererefersto replacementof crack pixel under observation by the median of the local observationsi,.e., equalto one, amongthe neighbouringpixels. The filled crack pixels are definedby:
y , = m e d ( x , _ j . . . . . x i . . . . .,*) ,
(3)
wherex arethe pixels in the local regionof the window andj: (N-l)/2 . For colour paintings the same process is usedon threeindependenct hannelsindividually and then combined to obtain crack filled colour paintings. Based on experimentation on various digitizedpaintings(both Gray and Colour) it has been found that our method gives results which can be judged qualitatively by observing restored paintings shown in Figs. 6(b), 7(b), 8(b) and 9(b) against the associated cracked paintings ( Figs.6(a), 7(a), 8(a) and 9(a)). Further, our method along with filling cracks is able to preservesharp edgesof the paintings as is evident from comparison of restoredpainting of Figs.lO (b) and ll(b) againstits counterpart in Figs. 10(a) and l1(a) respectively.This is verified usingCanny'sedgedetectiontechnique.
Figure 6 (a). Crackedpainting(b). Restored painting
69
ThammasatInt. Figure 8 (a). Crackedpainting(b). Restored painting Figure 7 (a). Crackedpainting(b). Restored painting Figure 9 (a). Crackedpainting(b). Restored painting 10
ThammasatInt. J. Sc. Tech..Vol. 13. No.3. Julv-Seotembe2r008
Figure l0 (a). Restoredpainting(b). Restored paintingedges
Figure I I (a). Restoredpainting(b). Restored paintingedges 4. Conclusion This paper presentsa new crack detection model employing BHT operation followed by thresholdingand MAO. This model producesa crackmap consistingof true cracksas illustrated in Figs.2 to 5. Further,we haveemployeda new filter MAMF (ModifiedAdaptiveMedianFilter) to fill in the thick and thin cracksas shownin Figs.6 to 9. Figs. l0 and I I highlightthe edges preservedby MAMF employedby us, which has been verified by the Canny edge detection technique. 5. References tll I. Giakoumis,N. Nikolaidis,and L Pitas, Digital lmage ProcessingTechniquesfor the Detection and Removal of Cracks in Digitized Paintings,IEEE Transactionson I m a g e P r o c e s s i n gV, o l . l 5 , N o .I , p p . l 7 8 l 88.Jan.2006. I2l F.S. Abas, and K. Martinez, Classification of Painting Cracks for Content-based Analysis, in Proc. of IS&T/SPIE's l5th Annual Symposiumon Electroniclmaging: Machine Vision Applications in Industrial I n s p e c t i o nV,o l . I l , p p . 1 4 9 - 1 6 0 , 2 0 0 3 . t3l M. Bami, F. Bartolini,and V. Cappellini, ImageProcessingfor Virtual Restorationof Artworks,IEEE. Multimedia,Vol.7, No. 2, p p . 3 4 - 3 7 ,J u n . 2 0 0 0 . l4l A. Hanbury, P. Kammerer, and E. Zolda, Painting Crack Elimination using Viscous Morphological Reconstruction, in Proc. IEEE l2th Int. Conf. Imase Analvsis and
'7 1
Thammasat
ProcessingM, antua, l7-19, pp. 226-231, S e p t .2 0 0 3 Ist I. Giakoumis and, I. Pitas, Digital Restoration of Painting cracks, in Proc. IEEE Int. Symp. Circuits and Systems, Monterey,Vol. 4, pp.269-272,1998, t6l L Pitas, and A. N. VenetsanoPoulos, Nonlinear digital filters, Principles and Applications,Norwel, MA: Kluwer, 1990.
17l L. Vincent, Morphological area Opening andClosingsfor GreyscaleImages,in Proc. Springer Verlag Shape in Picture NATO Workshop,Driebergen,Sept.1992. t8l N. Otsu, A Threshold Selection Method from Gray-level Histograms, IEEE Transactions on Systems, Man, CyberneticsV, ol. 9, No.l, pp. 62-66,1919.
12

A Gupta, V Khandelwal, A Gupta

File: image-processing-methods-for-the-restoration-of-digitized-paintings.pdf
Author: A Gupta, V Khandelwal, A Gupta
Published: Wed Oct 1 15:32:44 2008
Pages: 7
File size: 1.18 Mb


The Entrepreneurial City, 6 pages, 0.39 Mb

Art and the scalpel, 20 pages, 0.54 Mb

Elementary linear algebra, 7 pages, 0.44 Mb

How to think with your gut, 6 pages, 0.05 Mb
Copyright © 2018 doc.uments.com