Digital Art History

A Subject in Transition. Computers and the History of Art Series, Volume 1

Anna Bentkowska-Kafel, Trish Cashen, Hazel Gardiner

• Publisher: Intellect Books
• Year: 2005
• ISBN:
  9781841509105
• Edition: 1
• Topic: Art
• Subtopic: History of Art

This book looks at the transformation that Art and Art history is undergoing through engagement with the digital revolution. Since its initiation in 1985, CHArt (Computers and the History of Art) has set out to promote interaction between the rapidly developing new Information Technology and the study and practice of Art. It has become increasingly clear in recent years that this interaction has led, not just to the provision of new tools for the carrying out of existing practices, but to the evolution of unprecedented activities and modes of thought. This collection of papers represents the variety, innovation and richness of significant presentations made at the CHArt Conferences of 2001 and 2002. Some show new methods of teaching being employed, making clear in particular the huge advantages that IT can provide for engaging students in learning and interactive discussion. It also shows how much is to be gained from the flexibility of the digital image – or could be gained if the road block of copyright is finally overcome. Others look at the impact on collections and archives, showing exciting ways of using computers to make available information about collections and archives and to provide new accessibility to archives. The way such material can now be accessed via the internet has revolutionized the search methods of scholars, but it has also made information available to all. However the internet is not only about access. Some papers here show how it also offers the opportunity of exploring the structure of images and dealing with the fascinating possibilities offered by digitisation for visual analysis, searching and reconstruction. Another challenging aspect covered here are the possibilities offered by digital media for new art forms. One point that emerges is that digital art is not some discreet practice, separated from other art forms. It is rather an approach that can involve all manner of association with both other art practices and with other forms of presentation and enquiry, demonstrating that we are witnessing a revolution that affects all our activities and not one that simply leads to the establishment of a new discipline to set alongside others.

Bringing Pictorial Space to Life:

Computer Techniques for the Analysis of Paintings¹

Antonio Criminisi, Martin Kemp and Andrew Zisserman

1. Introduction

In the twentieth century, art and science were generally perceived as very diverse disciplines, with very few points of contact between them. Although there are signs that the schism is less sharp than it was, we are a long way from the situation that prevailed in the Italian Renaissance, when the distinction between those who practised what we call art and science was not sharply drawn.

The work in this paper – building upon that by Kemp² – aspires to show how scientific analysis and the study of art can interact and be mutually beneficial in achieving their goals. Novel and powerful computer techniques may help art historians to answer questions about geometry, depth and structure in Renaissance paintings. The focus of this paper is to show how computer graphics and computer vision can help give new kinds of answers to these and other questions about the spatial structure of paintings.

All the paintings that will be taken into consideration in this work share a great sense of perspective. Indeed, a perspectival structure of some elaboration is necessary if the proposed method is to yield meaningful results.

Filippo Brunelleschi’s mid-fifteenth-century invention of linear perspective was quickly adopted by contemporaries such as Masaccio, Donatello, Piero della Francesca, Domenico Veneziano and Paolo Uccello as the best technique to convey the illusion of a three-dimensional scene on a flat surface such as a panel or a wall. Later, in the seventeenth and eighteenth centuries mathematicians such as Desargues, Pascal, Taylor and Monge who were interested in linear perspective, laid the foundations of modern projective geometry.³ Projective geometry can be regarded as a powerful tool for modelling the rules of linear perspective in a metrical or algebraic framework.

Over the past ten years projective geometry has become the basis of many of the most powerful computer vision algorithms for three-dimensional reconstruction from multiple views.⁴ In particular, the work on single-view metrology⁵ provides tools and techniques to compute geometrically accurate three-dimensional models from single perspective images. These single-view techniques are applicable to all perspectival images (such as photographs) and are extensively applied, in this paper, to paintings that adhere to the canonical rules of linear perspective in order to create a systematic illusion of space behind the picture plane.

This paper, rather than just showing three-dimensional reconstructions designed on the basis of data extracted from paintings, presents novel and general methods that may be applied directly to any perspective image for a thorough analysis of its geometry.

We should constantly bear in mind that a painting is a creation that relies upon the artist’s and spectator’s imagination to construct a new or artificial world, one in which the manipulation of orthodox perspective may be advantageous. Therefore, any purely geometric analysis must be carried out in a sensitive manner. Before any geometric reconstruction is applied it is necessary to ascertain the level of geometric accuracy within the painting and, by implication, the desire of its maker for perspectival precision.

This paper presents some simple techniques for assessing the consistency of painted geometry. This is done by using powerful techniques to analyse the location of vanishing points and vanishing lines, by checking the symmetry of arches and other curved structures and by analysing the rate of diminution of receding patterns (e.g. tiled floors and arched vaults).

If a painting conforms to the rules of linear perspective then it behaves, geometrically, as a perspective image and it can be treated as analogous to a straightforward photograph of an actual subject. Vision algorithms can then be applied to: generate new views of the painted scenes; analyse shapes and proportions of objects in the scene; complete partially occluded objects; reconstruct missing regions of patterns; perform a complete 3D reconstruction and an analysis of the possible ambiguities in reconstruction.

1.1. Alternative Methods in History of Art

The method advocated here should be set in the context of previous techniques for the analysis of perspective in paintings. There are now three main alternatives that exhibit different sets of advantages and disadvantages.

Hand-made

The longest standing method has been to draw lines on the surface of paintings, or, rather, on the surfaces of reproductions of paintings (for obvious reasons). The linear analyses may be conducted either directly on a reproduction, or by using transparent overlays, and the results may be shown either superimposed on a reproduction or as separate diagrams. Kemp⁶ amongst others adopted this latter form of presentation. The analysis of a painting should preferably be performed on as large a reproduction as is available, ideally life-size (though this is rarely possible). The advantages of the hand-drawn analyses are:

• The technique is congruent with that used by most artists, who typically constructed their illusions of space through preparatory work involving straight edge and linear measures (sometimes also using dividers and compasses);

• In the initial stage of analysis, the hand-drawn lines can work flexibly for experimental exploration of our intuitions about the structure of the depicted space.

The disadvantages are:

• It is all too common to find thick lines drawn on small reproductions of big paintings, with resultant imprecision;

• It is easy to make errors, as when drawing a line through the point of intersection of two other lines, in which the extension of the resulting line will deviate progressively in response to the error in exact placement at the vertex of the intersecting lines.

• Constructing diagrams for complex illusions is a lengthy process.

Traditional Computer Aided Design

More recently, data obtained from the analysis of paintings has been used to obtain a computer aided design (CAD) reconstruction of the depicted space, using standard programs.⁷

The advantages of the CAD reconstructions are:

• They depict spatial features with precision according to the classic rules of linear perspective;

• Using techniques for rendering, they produce pictorial effects of light, shade, colour and (to a degree) texture akin to those in the original image;

• They can be used to produce animated fly-throughs and externalised views of the reconstructed spaces that can be vivid aids to understanding.

The disadvantages are:

• They require data to be extracted in advance from the painting, often in an artificially ‘tidied-up’ manner, in order to work with the program;

• They acquire a separate existence from the original images and may assume an aura of precision and conviction that is attractive but spurious.

The Present Method

The method advocated here still belongs to the wide spectrum of CAD applications, but unlike traditional techniques, works directly from the surface of the painting, and does not, for the purpose of analysis, add any arbitrary data not embedded in the image itself.

The advantages are:

• Alternative starting assumptions about the space in the image can be explored with ease and compared;

• All the alternatives can be parametised in a rigorous, mathematical fashion;

• The internal consistencies and inconsistencies of the spatial representation are laid bare, using information available directly from the image itself and involving no re-drawing;

• The textures and colours of the original image are retained;

• Re-projection can scrutinise errors and allow for their correction;

• Degrees of inaccuracy can be estimated systematically across the surface of images;

• Fly-throughs and externalised views of the space can be produced with ease;

• Regions occluded by objects closer to the viewer can be systematically reconstructed.

The disadvantages are:

• The power of the analysis may be excessive for the quality of the information that the artist entered into the original painting;

• It is applicable with profit only to those paintings that were constructed with sustained attention to perspectival rules;

• The quality of resolution is dependent upon the quality of the source image.

Some of the disadvantages of the present method may apply also to the previous two methods.

The technique described in this paper could be seen as a more flexible computer- assisted analytical tool which, unlike other CAD applications, allows the user to analyse the geometrical information contained in the painting in a more rigorous way by exploring all the possible reconstruction alternatives systematically, avoiding implicit assumptions typical of the use of traditional geometric templates.

It seems that the third method is generally superior for the analysis of complex perspectival images, and point to the first three advantages as empirically decisive in relation to the other two methods.

Correlating the Results with Historical Knowledge

The results obtained by these or other possible methods all need to be correlated by the historian with three other main bodies of evidence:

1. The archaeology of the painting; that is to say the physical evidence embedded in the work which reveals the constructional methods employed. These include incised lines (such as are apparent in Masaccio’s Trinity), underdrawings detectable with such techniques as infrared reflectography, and any pentimenti (changes of mind) visible in the surface of the painting;

2. Evidence from drawings by the artist and comparable artists about the methods they employed to construct perspectival spaces;

3. The techniques for spatial construction available at the time the paintings were made, as recorded in published and unpublished treatises and diagrams.

The remaining sections of the paper are organised as follows: section two describes some techniques to assess the accuracy of a painting’s linear perspective. Section three presents algorithms to analyse patterns and shapes of objects in paintings. Finally, section four presents complete three-dimensional reconstructions of paintings, analyses the dependency of the reconstructed geometry upon the assumptions made and explores the possible reconstruction ambiguities. Throughout the paper the analytical work of art historians is shown side by side with the results originated by applying our efficient and rigorous vision techniques to paintings.

2. Assessing the Accuracy of a Painting’s Linear Perspective

As stated in the introduction, injudicious application of reconstruction techniques to paintings may lead to disastrous results. Our first task is to assess how well a painting adheres to the rules of linear perspective; in other words, how a...