6533b82efe1ef96bd129389e

RESEARCH PRODUCT

A multilight approach for documenting and modeling the appearance of large cultural heritage objects

subject

description

In recent years, Computer Vision has played a major role in the investigation of sur-face appearance, particularly with the introduction of image-based rendering algorithms. Reflectance transformation Imaging (RTI) is one of these approaches. The RTI is a multi-light imaging technique that involves taking a series of images of an item using a stationary camera, while the location of the incident light varies in each photograph. Thanks to the information gathered from this type of acquisition, which is also defined as a Multi-light Image Collection (MLIC), it is possible to model the angular reflectance of a surface. The RTI is frequently employed in the documentation and study of architectural and cultural heritage objects, owing to its flexibility and convenience of usage, as it does not need the employmentof specialized and/or expensive equipment. However, this flexibility can occasionally have a negative impact on the quality of the visual representation achieved with this approach. Taking this into account, the primary goals of this research are to improve the RTI technique from both a methodological and an instrumental viewpoint, with a particular emphasis on the documentation and characterization of large works of art, as well as the study of multi-modal approaches. Thus, this work presents the study, development and implementation of two methodologies focused on the correction of artifacts related to the light source. In the first one, we investigate how the angular distribution of light source positions affects theRTI's reconstruction of surface reflectance. The proposed solution relies on estimating the local density of the light source positions distribution in order to weigh it and thus calibrate non-homogeneous distributions. The second focuses on the non-homogeneous illumination that is commonly present in RTI images, which is linked with the non-collimated lighting sources that are typically employed in the technique's application. As a solution to this problem, we present an approach that exploits the model of a near punctual light to obtain illumination correction at a pixel level. Finally, we delve into two understudied areas within the RTI framework, namely the acquisition of large objects and the expansion into the multi-spectral field. In this part, we offer methodology adaptation and improvement fromboth an instrumental and a methodological standpoint.