The concern and interest of this project is the non-destructive, contactless registration of 3D and multispectral data for the study of cultural heritage. 3D models record the 3D surface of the object under study and can be used to observe the exterior structure of an object without manipulating it. Conservators can thus easily examine brushstrokes on paintings and chisel marks of statues. Multispectral imaging, on the other hand, aims at recording the object‘s reflectance properties. Common applications include producing more faithful color reproductions, guiding pigment identification and improving readability of covered surfaces.
Combining these two methods provides 3D models with multispectral texture that are most useful for analysis purposes; highlighting spectral and structural correlations. However, when dealing with separate datasets, the mandatory registration of multispectral and 3D data is not an easy task, particularly when dealing with data describing cultural heritage. In this context, the use of targets is generally proscribed. Though it is possible, in some specific cases, to use targets to guide the registration, these defeat the purpose of using contactless acquisition systems.
Registration methods that rely on the structure of the data itself are also not adapted: the very fact that these techniques are complementary means that there may be few or no corresponding points between the multimodal datasets. Smooth surfaces covered with a flat tint, which are not unusual in paintings, present no salient point, be it 3D or multispectral. Despite these difficulties, we need a high registration accuracy, of the order of a fraction of a pixel, if these integrated models are to be used for analysis purposes.
The method was then used for the registration of 3D and multispectral datasets. The multispectral data is projected as a texture on a 3D model whose constituent meshes have been previously registered using our technique. The projection parameters are, as before, evaluated from the tracking of the target frame as well as from the optical and geometrical calibrations.
The 3D/multispectral registration was first tested on an object with much greater spectral and structural variability than typical cultural heritage objects, enabling a quick visual assessment of the accuracy of the final registration. We also applied our method to the study of an object with similar spatial and spectral variations as a real object we are interested in. These test cases permitted to asses the accuracy of our system, to highlight its limitations and globally to make sure that this method can be applied in situ with reasonable constraints. Throughout these tests, we improved the setup till we deemed it appropriate for the in situ study of cultural heritage.
Our method enables the registration of 3D and multispectral data describing any surface. The final registration accuracy is independent from the content of the acquired data and thus adapted to registering data with no salient features. The method is transportable and can be used in situ. Other optical sensors than the ones used in our study can be employed. This includes not only other 3D digitization systems and multispectral cameras, but also other optical sensors such as thermal sensors. The technique can thus be applied to other domains than cultural heritage, for example industrial metrology and quality control.