![Sample our Museum and Heritage Studies journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5945/TOC-Subject-Sample-2021-Museum-Heritage-Studies.jpg"})
Abstract
Surface texture is a critical feature in the manufacture, marketing, and use of photographic paper. Raking light reveals texture through a stark rendering of highlights and shadows. Though close-up raking light images effectively document surface features of photographic paper, the sheer number and diversity of textures used for historic papers prohibits efficient visual classification. This work provides evidence that automatic, computer-based classification of texture documented with raking light is feasible by demonstrating an encouraging degree of success sorting a set of 120 images made from samples of historic silver gelatin paper. Using this dataset, four university teams applied different image-processing strategies for automatic feature extraction and degree of similarity quantification. All four approaches successfully detected strong affinities and outliers built into the dataset. The creation and deployment of the algorithms was carried out by the teams without prior knowledge of the distributions of similarities and outliers. These results indicate that automatic classification of silver gelatin photographic paper based on close-up texture images is feasible and should be pursued. To encourage the development of other classification schemes, the 120-sample “training” dataset used in this work is available to other academic researchers at http://www.PaperTextureID.org.
Résumé – La texture de la surface du papier photographique est une caractéristique essentielle de sa fabrication, de sa commercialisation et de son utilisation. La lumière rasante révèle la texture du papier par un rendu saisissant des lumières et des ombres. Bien que les gros plans en lumière rasante documentent efficacement les caractéristiques de la surface des papiers photographiques, le nombre et la diversité des textures des papiers historiques empêchent une classification visuelle efficace. Le présent travail démontre que la classification automatique par ordinateur de textures documentées avec une lumière rasante est possible, en démontrant un degré encourageant de succès dans le tri d'un ensemble de 120 images réalisées à partir d'échantillons de papiers historiques au gélatino-bromure d'argent. À partir de ce groupe d'images, quatre équipes universitaires ont appliqué différentes stratégies de traitement d'image afin d'extraire automatiquement des caractéristiques et de quantifier le degré de similitude des papiers. Les quatre approches ont chacune réussi à détecter de fortes affinités, ainsi que des images avec caractéristique aberrante intégrées dans le groupe. La création et le déploiement des algorithmes ont été réalisés par les équipes sans connaissance préalable de la répartition des similitudes et des images aberrantes. Ces résultats indiquent que la classification automatique des papiers photographiques au gélatino-bromure d'argent basée sur des gros plans de texture est possible et doit être poursuivie. Pour encourager le développement d'autres systèmes de classification, les données des 120 échantillons utilisées dans ce travail sont mises à la disposition d'autres chercheurs scientifiques à l'adresse suivante: http://www.PaperTextureID.org.
Resumen – La textura superficial es un rasgo crítico en la manufactura, comercialización y uso de papel fotográfico. La luz rasante revela la textura a través de una representación nítida de áreas luminosas y sombras. A pesar de que las imágenes con luz rasante documentan efectivamente las características superficiales del papel fotográfico, la cantidad misma y la diversidad de texturas usadas en papeles históricos hace imposible una clasificación visual eficiente. Este trabajo proporciona evidencias de que la clasificación automática de la textura documentada con luz rasante, realizada por computadora es factible al demostrar un motivante grado de éxito al clasificar un grupo de 120 imágenes hechas de muestras de papel histórico de gelatina de plata. Usando esta base de datos, cuatro equipos universitarios aplicaron diferentes estrategias de procesamiento de imágenes para la extracción automática de rasgos y cuantificación del grado de similitud. Las cuatro estrategias detectaron con éxito afinidades fuertes así como los casos atípicos incluidos en la base de datos. Los equipos crearon y usaron algoritmos sin conocimiento previo de las similitudes y los casos lejanos al promedio. Estos resultados indican que la clasificación automática de papel fotográfico de gelatina de plata basada en imágenes que muestran acercamientos de la textura es viable y debería realizarse. Para incentivar el desarrollo de otros esquemas de clasificación el grupo de 120 muestras de “entrenamiento” usado en este trabajo está disponible para otros investigadores en http://www.PaperTextureID.org.
Resumo – Textura de superfície é um elemento crucial na fabricação, comercialização e uso de papéis fotográficos. Luz rasante revela a textura através da renderização acentuada de altas luzes e sombras. Ainda que imagens feitas com luz rasante em close-up efetivamente documentem as características da superfície dos papéis fotográficos, a enorme quantidade e diversidade de texturas utilizadas nos papéis históricos impedem a classificação visual eficiente. Esse trabalho fornece evidências de que a classificação de texturas documentadas com luz rasante em base de computador, automatizada, é viável pela demonstração bem sucedida e promissora da ordenação de um conjunto de 120 imagens feitas de amostras de papel histórico de gelatina e prata. Com esse conjunto de dados, equipes de quatro universidades aplicaram estratégias diferentes de processamento de imagem para extração automatizada de características e quantificação de graus de semelhança. Todas as quatro abordagens detectaram com sucesso relevantes afinidades e variações atípicas (desvios-padrão) incorporadas ao conjunto de dados. A criação e implantação de algoritmos foram realizadas por equipes sem conhecimento prévio das distribuições de similaridades e de desvios-padrão. Os resultados indicaram que a classificação automatizada de papéis fotográficos de gelatina e prata fundamentada nas imagens em close-up das texturas é possível e deve ser alcançada. Para estimular o desenvolvimento de outros esquemas de classificação, o conjunto “teste” de dados com 120 amostras utilizado nesse trabalho está disponível para outras pesquisas acadêmicas no http://www.PaperTextureID.org.
Acknowledgments
The authors wish to thank Jill Sterret and Theresa Andrews of the San Francisco Museum of Modern Art for providing a meeting venue during the summer of 2012, Andrew Messier, Lincoln Laboratories, Massachusetts Institute of Technology, for help designing the imaging system, and Ian Holland, Lumenera Technical Assistance Center, for assistance with imaging specifications.
Glossary
2D-Discrete Wavelet Transform (2D-DWT). The 2D-DWT is a classical image processing tool that provides a multiscale, exact, and invertible representation of an image by assessing the collection of image versions band-pass filtered at different scales.
Cepstral distance. The word cepstrum was invented in 1963 and derives from a reordering of the first letters in spectrum. A cepstrum is the result of analyzing signals or images with (a normalized version of) the logarithm of the absolute value of a classical spectrum, so as to better permit comparisons in terms of change rate, magnitude, phase, power, and other features. The cepstral distance is the net result of this comparison.
Eigentextures. For a given image, eigentextures are the columns of an imposed matrix that form the basis for a set of all depicted textures. The eigentextures provide a rank ordering of the most significant of these sub-spaces, thus providing a reduced dimensional approximation, or distillation, of the texture pattern.
k-means clustering. k-means clustering is an adapted signal-processing technique used to analyze vectors and other complex variables for the purposes of modeling large datasets and to derive affinity groupings. k-means clustering is often used for data mining and machine learning applications.
Least Squares. Least squares is a standard method for data fitting and regression analysis. Data fitting refers to the determination of a curve or mathematical function that best describes (or “fits”) a set of data points according to a particular measure, i.e. the sum of squared errors to each data point. Similarly, regression analysis is used to estimate the relationships between variables.
Multiscale. Multiscale methods refer to a broad range of signal processing and classification techniques that analyze data jointly or simultaneously at different scales or resolutions. Often, multiscale techniques rely on fitting and comparing data against exemplar structural building blocks, also at different scales.
Non-semantic. Non-semantic methods also refer to a broad range of signal processing and classification techniques. As opposed to multiscale techniques, non-semantic methods are based on values measured directly from the sample under consideration versus the application of prototypical patterns and data structures.
Hyperbolic wavelet transform (HWT): The hyperbolic wavelet transform extends the 2D-DWT by allowing the use of different dilation factors (changes of scale) on the horizontal and vertical axes of a data matrix. Image analysis using HWT allows both multiscale and anisotropic feature extraction.
Singular value decomposition. Singular value decomposition is a mathematical method for a data-driven derivation of substructures, patterns, correlations, and variations from complex, multi-dimensional data matrices.
Textons. Textons are the most basic texture elements that, when repeated, fully define an image depicting a textured surface. These fundamental micro-structures are often described as the “atoms” of texture perception.
Additional information
Notes on contributors
C. Richard Johnson
Paul Messier is an independent conservator of photographs working in Boston, Massachusetts. Founded in 1994, his studio provides conservation services for private and institutional clients throughout the world. The heart of this practice is unique knowledge and ongoing research into photographic papers. The Messier Reference Collection of Photographic Papers plays a vital role in this work. He received a Masters of Arts and certificate of advanced study in the conservation of works on paper from the art conservation program at the State University College at Buffalo (SUNY). Messier is the corresponding author. Address: 103 Brooks Street, Boston, MA 02135. Email: [email protected] A full list of author biographies for this paper can be found at http://www.maneyonline.com/doi/suppl/10.1179/1945233014Y.0000000024.S1
Paul Messier
C. Richard Johnson, Jr. received a PhD in Electrical Engineering from Stanford University, along with the first PhD minor in Art History granted by Stanford, in 1977. He is currently the Geoffrey S. M. Hedrick Senior Professor of Engineering and a Stephen H. Weiss Presidential Fellow at Cornell University. Since 2005, his primary research interest has been computational art history. Professor Johnson founded the Thread Count Automation Project (TCAP) in collaboration with the Van Gogh Museum in 2007, initiated the Historic Photographic Paper Classification (HPPC) challenge in cooperation with the Museum of Modern Art in 2010, and launched the Chain Line Pattern (CLiP) Matching Project with the Morgan Library & Museum in 2012, with the Rijksmuseum and the Metropolitan Museum of Art joining the project in 2013. He was an Adjunct Research Fellow of the Van Gogh Museum from 2007 through 2011. In 2013, Professor Johnson was appointed a Scientific Researcher of the Rijksmuseum and Computational Art History Advisor to the RKD – Netherlands Institute for Art History. Address: School of Electrical and Computer Engineering, 390 Rhodes Hall, Cornell University, Ithaca, NY 14853. Email: [email protected]