1. The Extraction and Classification of Craquelure for Geographical and Conditional Based Analysis of Pictorial Art Mouhanned El-Youssef, Spike Bucklow, Roman Gr. Maev

2. Background Craquelure represents the unique crack formations on paintings that is related to the materials and methods employed by the artist Paintings from a particular geographical area tend to exhibit similar trends in their craquelure patterns Craquelure is of important interest to art conservators for they indicate the condition of the painting and areas potentially exhibiting mechanical stress

3. Background Image courtesy of Dr. Bucklow French Painting – Smooth crack lines – Curved – Large Islands

4. Background Image courtesy of Dr. Bucklow Flemish Painting – Jagged crack lines – Ordered and straight – Small Islands

5. Objectives To develop a craquelure analysis algorithm capable of classifying each craquelure pattern to their corresponding region of origins To establish a crack monitoring system that will monitor and track the growth of cracks over the course of a painting’s life

6. Agenda 1)Project Outline 2)Crack Extraction Techniques 3)Crack Hierarchal Representation 4)Feature Extraction/Dimension Reduction 5)Classification 6)Crack Monitoring System 7)Questions and Discussion

7. Project Outline

8. Crack Extraction Techniques Mathematical Morphology is used to extract crack lines from an image Mathematical Morphology consists of four operations: – Dilation – Erosion – Closing (Dilation Erosion) – Opening (Erosion Dilation)

9. Flash Video

10. Mathematical Morphology

11. Crack Extraction Techniques: Thresholding Global thresholding Segmented thresholding Offset Thresholding The greyscale image then needs to be thresholded to a binary image. Three general thresholding techniques:

12. Global Thresholding

13. Segmented Thresholding

14. Offset Thresholding

15. Comparison Global ThresholdSegmented ThresholdOffset Threshold

16. Crack Hierarchal Representation After crack extractions, a line following algorithm is applied to the image A sequence of numbers (chain code) is recorded based on the crack direction Edge points, node points and line lengths are determined in a hierarchal fashion

17. Crack Hierarchal Representation An illustration of the crack following algorithm is displayed below: Crack Following Algorithm: Each neighbouring pixel is assigned a number value The crack line above would display the following chain code: 258225578

18. Crack Hierarchal Representation After applying the crack following algorithm, our crack pattern is structured in terms of nodes, edges, lines and distances The green lines in the figure are crack lines that form an island (a closed loop). Red pixels represent nodes and yellow pixels represent edge points.

19. Feature Extraction The length of each individual line segment is calculated during the hierarchal representation of the crack pattern Various other features are extracted from the network A histogram is obtained by tallying the direction vectors in the chain code The number of islands is calculated within a crack network

20. Feature Extraction Overall features extracted are: Number of Islands Node density Edge density Directionality Histogram Distance between nodes Straight line to actual line ratio Line Jaggedness Lines per node Average line length

21. Feature Extraction Line size is calculated as: Straight line to actual line ratio: Line Significance:

22. Feature Extraction Flemish Craquelure Italian Craquelure Dutch Craquelure French Craquelure

23. Dimension Reduction Reducing the dimensions of a feature space is necessary before classification Principle Component Analysis (PCA) as well as Linear Discriminant Analysis (LDA) is used for feature reduction The Fisher ratio test is used to determine which variable best discriminates between classes

24. Classification After dimension reduction, the features were fed into a Linear Discriminatory Analysis (LDA) classifier for classification The results were found to be: Predicted Class ClassFlemishFrenchItalianDutch Actual Class Flemish67.0%0%12.5%20.5% French0%92.5%6.0%1.5% Italian22.5%10%59.0%8.5% Dutch34%0%6.0%60.0%

25. Crack Monitoring System The crack monitoring system is used to monitor the growth of cracks on the surface to determine the “health” of a painting. The monitoring system must be: Must be rotation invariant Must be scale invariant Must be robust

26. Crack Monitoring System Present1 year ago

27. Crack Monitoring System

29. Conclusions A framework for the classification of craquelure patterns has been established A health monitoring system can be used to detect and monitor the growth of cracks in paintings

30. Future Work Expand on the classification process by investigating alternative techniques Continue building a database of craquelure images for different types of paintings (ie. French, Dutch..) Refine the health monitoring algorithm and develop a set of acquisition constraints

31. Questions and Discussion

32. Refrences [1] Spike Bucklow, "The Description of Craquelure Patterns," Studies in Conservation, vol. 42, no. 3, pp. 129-140, 1997. [2] Alexandra Moxley, "The Art of Fogery," The Undergraduate Scholar, pp. 1-10, 2010. [3] Spike Bucklow, "A Stylometric Analysis of Craquelure," Computers and the Humanities, vol. 31, pp. 503-521, 1998. [4] Fazly Salleh Abas, "Analysis of Craquelure Patterns for Content-Based Retrieval," University of Southampton, Southampton, PhD Thesis 2004. [5] Irene Crisologo, Chrisopher Monterola, and Maricor Soriano, "Statistical Feature- based Craquelure Classification," International Journal of Modern Physics C, vol. 22, no. 11, pp. 1191-1209, September 2011. [6] P. Willigen, A Mathematical Study on Craquelure and other Mechanical Damage in Paintings, E Klerk and H. Maaren, Eds. Dellf, The Netherlands: Delft Univeristy Press, 1999. [7] W. Stanley Taft and James W. Mayer, The Science of Paintings. New York, USA: Springer-Verlag, 2000.