This PDF file contains the front matter associated with SPIE Proceedings Volume 7869, including the Title Page, Copyright information, Table of Contents, and the Conference Committee listing.

Stylometry in visual art-the mathematical description of artists' styles - has been based on a number of properties of works, such as color, brush stroke shape, visual texture, and measures of contours' curvatures. We introduce the concept of quantitative measures of lighting, such as statistical descriptions of spatial coherence, diuseness, and so forth, as properties of artistic style. Some artists of the high Renaissance, such as Leonardo, worked from nature and strove to render illumination "faithfully" photorealists, such as Richard Estes, worked from photographs and duplicated the "physics based" lighting accurately. As such, each had dierent motivations, methodologies, stagings, and "accuracies" in rendering lighting clues. Perceptual studies show that observers are poor judges of properties of lighting in photographs such as consistency (and thus by extension in paintings as well); computer methods such as rigorous cast-shadow analysis, occluding-contour analysis and spherical harmonic based estimation of light fields can be quite accurate. For this reasons, computer lighting analysis can provide a new tools for art historical studies. We review lighting analysis in paintings such as Vermeer's Girl with a pearl earring, de la Tour's Christ in the carpenter's studio, Caravaggio's Magdalen with the smoking flame and Calling of St. Matthew) and extend our corpus to works where lighting coherence is of interest to art historians, such as Caravaggio's Adoration of the Shepherds or Nativity (1609) in the Capuchin church of Santa Maria degli Angeli. Our measure of lighting coherence may help reveal the working methods of some artists and in diachronic studies of individual artists. We speculate on artists and art historical questions that may ultimately profit from future renements to these new computational tools.

The study of the visual art of printmaking is fundamental for art history. Printmaking methods have been used for centuries to replicate visual art works, which have influenced generations of artists. Particularly in this work, we are interested in the influence of prints on artistic tile panel painters, who have produced an impressive body of work in Portugal. The study of such panels has gained interest by art historians, who try to understand the influence of prints on tile panels artists in order to understand the evolution of this type of visual arts. Several databases of digitized art images have been used for such end, but the use of these databases relies on manual image annotations, an effective internal organization, and an ability of the art historian to visually recognize relevant prints. We propose an automation of these tasks using statistical pattern recognition techniques that takes into account not only the manual annotations available, but also the visual characteristics of the images. Specifically, we introduce a new network link-analysis method for the automatic annotation and retrieval of digital images of prints. Using a database of 307 annotated images of prints, we show that the annotation and retrieval results produced by our approach are better than the results of state-of-the-art content-based image retrieval methods.

Painted tile panels (Azulejos) are one of the most representative Portuguese forms of art. Most of these panels are inspired on, and sometimes are literal copies of, famous paintings, or prints of those paintings. In order to study the Azulejos, art historians need to trace these roots. To do that they manually search art image databases, looking for images similar to the representation on the tile panel. This is an overwhelming task that should be automated as much as possible. Among several cues, the pose of humans and the general composition of people in a scene is quite discriminative. We build an image descriptor, combining the kinematic chain of each character, and contextual information about their composition, in the scene. Given a query image, our system computes its similarity profile over the database. Using nearest neighbors in the space of the descriptors, the proposed system retrieves the prints that most likely inspired the tiles' work.

The aim of image understanding, which is a long standing goal of computer vision, is to develop algorithms with which computers can advance to the semantic content of images. One ability of such algorithms would be the automatic discovery of relations between different objects in large collections of images. To analyze this relatedness we present an unsupervised and a semi-supervised approach for decomposing the large intra-class variability of object categories. The relations between objects is discovered by mapping all exemplars into a single low-dimensional projection that preserves the structure that is inherent to the category. The analysis reveals subtypes and an automatic classification algorithm is presented that predicts the artistic workshop that has drawn the objects. Finally, an approach for ordering the instances of an object category is proposed that also shows transitions between object instances. Our work is based on late medieval manuscripts from the Codices Palatini germanici.

For many years, a lot of museums and countries organize the high definition digitalization of their own collections. In consequence, they generate massive data for each object. In this paper, we only focus on art painting collections. Nevertheless, we faced a very large database with heterogeneous data. Indeed, image collection includes very old and recent scans of negative photos, digital photos, multi and hyper spectral acquisitions, X-ray acquisition, and also front, back and lateral photos. Moreover, we have noted that art paintings suffer from much degradation: crack, softening, artifact, human damages and, overtime corruption. Considering that, it appears necessary to develop specific approaches and methods dedicated to digital art painting analysis. Consequently, this paper presents a complete framework to evaluate, compare and benchmark devoted to image processing algorithms.

Our paper addresses problem of multi-modal data acquisition and the following data visualization for art analysis and interpretation. Various types of modalities for acquisition of digital images are used for art analysis. The data we can obtain using various modalities differ in two ways. The acquired images can differ by their mutual geometry and possibly by their radiometric quality. These are the differences we would like to remove. The group of differences we are interested in are details or characteristics of an artwork, which are apparent just in the certain modality and which bring us new information. The two listed groups represent two categories of image processing methods we have to deal with. The first one is represented by image preprocessing methods such as data enhancement and restoration algorithms, the second class includes effective ways how to combine the acquired information into one image - image fusion. In our paper we present image quality enhancement for microscopic multi-modal data and their segmentation and recent results in data fusion and visualization for art analysis are demonstrated from the second category of methods.

During his explorations of Africa, David Livingstone kept a diary and wrote letters about his experiences. Near the end of his travels, he ran out of paper and ink and began recording his thoughts on leftover newspaper with ink made from local seeds. These writings suffer from fading, from interference with the printed text and from bleed through of the handwriting on the other side of the paper, making them hard to read. New image processing techniques have been developed to deal with these papers to make Livingstone's handwriting available to the scholars to read. A scan of the David Livingstone's papers was made using a twelve-wavelength, multispectral imaging system. The wavelengths ranged from the ultraviolet to the near infrared. In these wavelengths, the three different types of writing behave differently, making them distinguishable from each other. So far, three methods have been used to recover Livingstone's handwriting. These include pseudocolor (to make the different writings distinguishable), spectral band ratios (to remove text that does not change), and principal components analysis (to separate the different writings). In initial trials, these techniques have been able to lift handwriting off printed text and have suppressed handwriting that has bled through from the other side of the paper.

This paper addresses the problem of automating analyses of historical maps. The problem is motivated by the lack of accuracy and consistency in the current comparison process of geographical objects found in historical maps by visual inspections. The objective of our work is to compare shape characteristics of the Great Lakes region in a dataset of approximately 40 French and British historical maps created in the 17th through the 19th centuries. Our approach decomposes the visual inspection into steps such as object segmentation, spatial scale calibration, extraction of calibrated object descriptors and comparison of descriptors over time and multiple cartographer houses. The automation of object segmentation is achieved by template shape-based segmentation using the Hu moments as shape descriptors and ball-based region growing. The automation of spatial calibration is accomplished by detection and classification of lines along map borders and by mapping striped boundaries intersected by latitude and longitude lines into degrees of arc length. Thus, shape characteristics of segmentation results in pixels can be converted to geographical units, for example, an area of a lake in square miles. We report experimental evaluations of automation accuracy based on comparison with manual segmentation results, as well as the knowledge obtained from the area comparisons.

This paper shows the simulations of the usage of a LED cluster as the illumination source for a multispectral imaging system covering the range of wavelengths from 350 to 1650 nm. The system can be described as being composed of two modules determined by the spectral range of the imaging sensors responses, one of them covering the range from 350- 950nm (CCD camera) and the other one covering the wavelengths from 900-1650nm (InGaAs camera). A well known method of reflectance estimation, the pseudo-inverse method, jointly with the experimentally measured data of the spectral responses of the cameras and the spectral emission of the LED elements are used for the simulations. The performance of the system for spectral estimation under ideal conditions and realistic noise influence is evaluated through different spectral and colorimetric metrics like the GFC, RMS error and CIEDE2000 color difference formula. The results show that is expectable a rather good performance of the real setup. However, they also reveal a difference in the performances of the modules. The second module has poorer performance due to the less narrow spectral emission and less number of LED elements that covers the near-infrared spectral range.

As high-resolution images of paintings, acquired using various imaging modalities (e.g. X-ray, luminescence, visible and infrared reflection) become more available, it is increasingly useful to have accurate registration between them. Accurate registration allows new information to be compiled from the several multimodal images. This leads to a better understanding of how the painting was constructed and of any compositional changes that have occurred. To that end, we have produced an automatic image registration algorithm that is capable of aligning X-ray, color, and infrared images, as well as multispectral luminescence and reflectance image sets, or cubes. The key steps of the algorithm include identifying large sets of candidate control points in the reference image, then pairing them with potential points in a second image using cross-correlation. Finally, after selecting the best set of control point pairs, the second image is transformed to be in register with the reference image. Tests show the algorithm to be capable of achieving sub-pixel registration across these various image modalities.

Currently, a lot of different 3D scanning devices are used for 3D acquisition of art artifact surface shape and color. Each of them has different technical parameters starting from measurement principle (structured light, laser triangulation, interferometry, holography) and ending on parameters like measurement volume size, spatial resolution and precision of output data and color information. Some of the 3D scanners can grab additional information like surface normal vectors, BRDF distribution, multispectral color. In this paper, the problem of establishing of threshold for technical parameters of 3D scanning process as a function of required information about the object is discussed. Only two main technical parameters are under consideration, in order to cover as many different 3D scanning devices as possible - measurement sampling density (MSD - represented by number of points per square millimeter) and measurement uncertainty (MU - directly influencing final data accuracy). Also different materials and finishing techniques require different thresholds of MSD and MU parameters to collect similar documentation (for example documentation of object state for art conservation department) of different objects. In this paper we consider exemplary painting on canvas, wallpainting, graphics prints and stone samples to visualize what object features can be observed within different values of MSD and MU parameters.

Pigment degradation has been a subject of interest among researchers in the field of cultural heritage studies. Knowing how pigments behave when subjected to different elements such as high temperature, humidity, electromagnetic radiation and many more others is of prime importance. In this study, the effects of subjecting Japanese pigments to high temperature were investigated. Focus was given on the effects in terms of pigment discoloration and the micromechanism of degradation. Multispectral images were used to track the changes in color and spectral reflectance by reconstructing colorimetric and spectral information from the images. The multispectral images were taken using a high-resolution flat-bed scanner equipped with a line-CMOS camera. In addition, the pigments were characterized using commercially available spectrometers, X-ray diffraction, X-ray fluorescence spectroscopy and X-ray absorption fine structure were used to ascertain the influence of high temperature exposure of the pigments. The high resolution multispectral scans gave the most valuable insights into the discoloration and micromechanism of pigment degradation since they provide both analytical and visual information.

We derive and demonstrate new methods for dewarping images depicted in convex mirrors in artwork and for estimating the three-dimensional shapes of the mirrors themselves. Previous methods were based on the assumption that mirrors were spherical or paraboloidal, an assumption unlikely to hold for hand-blown glass spheres used in early Renaissance art, such as Johannes van Eyck's Portrait of Giovanni (?) Arnolfini and his wife (1434) and Robert Campin's Portrait of St. John the Baptist and Heinrich von Werl (1438). Our methods are more general than such previous methods in that we assume merely that the mirror is radially symmetric and that there are straight lines (or colinear points) in the actual source scene. We express the mirror's shape as a mathematical series and pose the image dewarping task as that of estimating the coefficients in the series expansion. Central to our method is the plumbline principle: that the optimal coefficients are those that dewarp the mirror image so as to straighten lines that correspond to straight lines in the source scene. We solve for these coefficients algebraically through principal component analysis, PCA. Our method relies on a global figure of merit to balance warping errors throughout the image and it thereby reduces a reliance on the somewhat subjective criterion used in earlier methods. Our estimation can be applied to separate image annuli, which is appropriate if the mirror shape is irregular. Once we have found the optimal image dewarping, we compute the mirror shape by solving a differential equation based on the estimated dewarping function. We demonstrate our methods on the Arnolfini mirror and reveal a dewarped image superior to those found in prior work|an image noticeably more rectilinear throughout and having a more coherent geometrical perspective and vanishing points. Moreover, we find the mirror deviated from spherical and paraboloidal shape; this implies that it would have been useless as a concave projection mirror, as has been claimed. Our dewarped image can be compared to the geometry in the full Arnolfini painting; the geometrical agreement strongly suggests that van Eyck worked from an actual room, not, as has been suggested by some art historians, a "fictive" room of his imagination. We apply our method to other mirrors depicted in art, such as Parmigianino's Self-portrait in a convex mirror and compare our results to those from earlier computer graphics simulations.

Time and order are considered crucial information in the art domain, and subject of many research efforts by historians. In this paper, we present a framework for estimating the ordering and date information of paintings and drawings. We formulate this problem as the embedding into a one dimension manifold, which aims to place paintings far or close to each other according to a measure of similarity. Our formulation can be seen as a manifold learning algorithm, albeit properly adapted to deal with existing questions in the art community. To solve this problem, we propose an approach based in Laplacian Eigenmaps and a convex optimization formulation. Both methods are able to incorporate art expertise as priors to the estimation, in the form of constraints. Types of information include exact or approximate dating and partial orderings. We explore the use of soft penalty terms to allow for constraint violation to account for the fact that prior knowledge may contain small errors. Our problem is tested within the scope of the PrintART project, which aims to assist art historians in tracing Portuguese Tile art "Azulejos" back to the engravings that inspired them. Furthermore, we describe other possible applications where time information (and hence, this method) could be of use in art history, fake detection or curatorial treatment.

Early attempts at authentication Jackson Pollock's drip paintings based on computer image analysis were restricted to a single "fractal" or "multi-fractal" visual feature, and achieved classification nearly indistinguishable from chance. Irfan and Stork pointed out that such Pollock authentication is an instance of visual texture recognition, a large discipline that universally relies on multiple visual features, and showed that modest, but statistically significant improvement in recognition accuracy can be achieved through the use of multiple features. Our work here extends such multi-feature classification by training on more image data and images of higher resolution of both genuine Pollocks and fakes. We exploit methods for feature extraction, feature selection and classiffier techniques commonly used in pattern recognition research including Support Vector Machines (SVM), decision trees (DT), and AdaBoost. We extract features from the fractality, multifractality, pink noise patterns, topological genus, and curvature properties of the images of candidate paintings, and address learning issues that have arisen due to the small number of examples. In our experiments, we found that the unmodified classiffiers like Support Vector Machines or Decision Tree alone give low accuracies (60%), but that statistical boosting through AdaBoost leads to accuracies of nearly 75%. Thus, although our set of observations is very small, we conclude that boosting methods can improve the accuracy of multi-feature classiffication of Pollock's drip paintings.

Underdrawings and pentimenti-typically revealed through x-ray imaging and infrared reflectography-comprise important evidence about the intermediate states of an artwork and thus the working methods of its creator.¹ To this end, Shahram, Stork and Donoho introduced the De-pict algorithm, which recovers layers of brush strokes in paintings with open brush work where several layers are partially visible, such as in van Gogh's Self portrait with a grey felt hat.² While that preliminary work served as a proof of concept that computer image analytic methods could recover some occluded brush strokes, the work needed further refinement before it could be a tool for art scholars. Our current work makes several steps to improve that algorithm. Specifically, we refine the inpainting step through the inclusion of curvature-based constraints, in which a mathematical curvature penalty biases the reconstruction toward matching the artist's smooth hand motion. We refine and test our methods using "ground truth" image data: passages of four layers of brush strokes in which the intermediate layers were recorded photographically. At each successive top layer (currently identified by the user), we used k-means clustering combined with graph cuts to obtain chromatically and spatially coherent segmentation of brush strokes. We then reconstructed strokes at the deeper layer with our new curvature-based inpainting algorithm based on chromatic level lines. Our methods are clearly superior to previous versions of the De-pict algorithm on van Gogh's works giving smoother, natural strokes that more closely match the shapes of unoccluded strokes. Our improved method might be applied to the classic drip paintings of Jackson Pollock, where the drip work is more open and the physics of splashing paint ensures that the curvature more uniform than in the brush strokes of van Gogh.

We examine one class of evidence put forth in support of the recent claim that the Italian Baroque master Caravaggio secretly employed optical projectors as a direct drawing aid. Specically, we test the claims that there is an "abnormal number" of left-handed gures in his works and, more specically, that "During the Del Monte period he had too many left-handed models." We also test whether there was a reversal in the handedness of specic models in different paintings. Such evidence would be consistent with the claim that Caravaggio switched between using a convex-lens projector to using a concave-mirror projector and would support, but not prove, the claim that Caravaggio used optical projections. We estimate the parity (+ or -) of each of Caravaggio's 76 appropriate oil paintings based on the handedness of gures, the orientation of asymmetric objects, placement of scabbards, depicted text, and so on, and search for statistically significant changes in handedness in figures. We also track the direction of the illumination over time in the artist's uvre. We discuss some historical evidence as it relates to the question of his possible use of optics. We nd the proportion of left-handed figures lower than that in the general population (not higher), and no significant change in estimated handedness even of individual models. Optical proponents have argued that Bacchus (1597) portrays a left-handed gure, but we give visual and cultural evidence showing that this gure is instead right-handed, thereby rebutting this claim that the painting was executed using optical projections. Moreover, scholars recently re-discovered the image of the artist with easel and canvas reflected in the carafe of wine at the front left in the tableau in Bacchus, showing that this painting was almost surely executed using traditional (non-optical) easel methods. We conclude that there is 1) no statistically signicant abnormally high number of left-handed gures in Caravaggio's uvre, including during any limited working period, 2) no statistically significant change in handedness among all gures or even individual gures that might be consistent with a change in optical projector, and 3) the visual and cultural evidence in Bacchus shows the gure was right-handed and that the artist executed this work by traditional (non-optical) easel methods. We conclude that the general parity and handedness evidence does not support the claim that Caravaggio employed optical projections.

David Hockney has argued that the right hand of the disciple, thrust to the rear in Caravaggio's Supper at Emmaus (1606), is anomalously large as a result of the artist refocusing a putative secret lens-based optical projector and tracing the image it projected onto his canvas. We show through rigorous optical analysis that to achieve such an anomalously large hand image, Caravaggio would have needed to make extremely large, conspicuous and implausible alterations to his studio setup, moving both his purported lens and his canvas nearly two meters between "exposing" the disciple's left hand and then his right hand. Such major disruptions to his studio would have impeded -not aided- Caravaggio in his work. Our optical analysis quantifies these problems and our computer graphics reconstruction of Caravaggio's studio illustrates these problems. In this way we conclude that Caravaggio did not use optical projections in the way claimed by Hockney, but instead most likely set the sizes of these hands "by eye" for artistic reasons.

Underdrawings and pentimenti reveal intermediate states of a painting and thus may shed light on the working methods of some artists. It has been claimed that Lorenzo Lotto used optical projections during the execution of Husband and wife (1543) and, recently, that underdrawings in that work might reveal evidence of tracing of optical projections. We analyze x-ray images of this painting-captured under careful, museum-laboratory conditions and enhanced through digital image processing-with special attention to the possibility of evidence of the use of optical projections in the central passage of the depicted carpet. We also study the work in situ and in high-resolution macro optical images of the central portion of the carpet pattern. These photographs reveal that the top portion of the keyhole pattern is not "blurry, like an out-of-focus image," but instead was merely executed in a somewhat broader brush than was neighboring passages. Furthermore, x-ray, infra-red and visible light inspection show that the white portions and black contours were executed atop a broad layer of dark red and reveal no record of an optical projection would have been present when Lotto executed the visible portion. As such, an evidence of putative underdrawings in this region has no bearing on the optical projection claim. There is no evidence of tracing marks-in charcoal or in any medium-in the top, visible portion of this passage either. As such, this visual, infra-red and x-ray evidence does not support the claim that this painting was executed under optical projections. We also discuss the difficulties with the projection theory with special reference to Lotto's preparatory drawing in the Rijksmuseum-specifically the need for a needlessly complex optical system (two lenses rather than one). We also review briefly contemporary textual evidence in early 16th-century Venice that has been used to support the optical projection claim for Lotto and conclude that it also fails to support the projection claim for this painting.

This work addresses the problem of automatic classification and labeling of 19th- and 20th-century quilts from photographs. The photographs are classified according to the quilt patterns into crazy and non - crazy categories. Based on the classification labels, humanists try to understand the distinct characteristics of an individual quilt-maker or relevant quilt-making groups in terms of their choices of pattern selection, color choices, layout, and original deviations from traditional patterns. While manual assignment of crazy and non-crazy labels can be achieved by visual inspection, there does not currently exist a clear definition of the level of crazy-ness, nor an automated method for classifying patterns as crazy and non-crazy. We approach the problem by modeling the level of crazy-ness by the distribution of clusters of color-homogeneous connected image segments of similar shapes. First, we extract signatures (a set of features) of quilt images that represent our model of crazy-ness. Next, we use a supervised classification method, such as the Support Vector Machine (SVM) with the radial basis function, to train and test the SVM model. Finally, the SVM model is optimized using N-fold cross validation and the classification accuracy is reported over a set of 39 quilt images.

Numerous cultural heritage art works have shiny surfaces resulting form gold, silver, and other metallic pigments. In addition varnish overlayer on oil paintings makes it challenging to retrieve true color information. This is due to the great effect of lighting condition when images are acquired and viewed. The reflection of light from such surfaces is a combination of the surface's specular and diffused light reflections. In this paper, this specific problems encountered when digitizing cultural heritage were discussed. Experimental results using the images acquired with a high-resolution large flat bed scanner, together with a mathematical method for processing the captured images were presented and discussed in detail. Focus was given in separating the diffused and specular components of the reflected light for the purpose of analytical imaging. The mathematical algorithm developed in this study enables imaging of cultural heritage with shiny and glossy surfaces effectively and efficiently.

Completely non-invasive digital cleaning of Fernando Amorsolo's 1948 oil on canvas, Malacañang by the River, is implemented using a trained neural network. The digital cleaning process results to more vivid colors and a higher luminosity for the digitally-cleaned painting. We propose three methods for visualizing the color change that occurred to a painting image after digital cleaning. For the first two visualizations, the color change between original and digitally-cleaned image is computed as a vector difference in RGB space. For the first visualization, the vector difference is projected on a neutral color and rendered for the whole image. The second visualization renders the color change as a translucent dirt layer that can be superimposed on a white image or on the digitally-cleaned image. For the third visualization, we model the color change as a dirt layer that acts as a filter on the painting image. The resulting color change and dirt layer visualizations are consistent with the actual perceived color change and could offer valuable insights to a painting's color changing process due to exposure.