380 
Lerma 
The overall classification accuracy achieved was 92.9 percent. The producer's accuracy for five classes were around or exceed 91 
percent; the producer's accuracy for red stone and pigment classes were around 85 percent, mainly due to the similarity (and finally 
interference) between both classes. 
The same number of classes achieved a user's accurary better than 94 percent, only brown stone and dark stone classes were below 
90 percent; the commission error in brown stone was the highest one, around 20%, most of it afecting unfortunately the pigment 
class. However, user's accuracy with pigment class was 93.1 percent. Both classes, firstly pigment and secondly brown stone were 
required for restoration tasks: they were properly defining and extracting the original pigmentation. 
It should be pointed out that visible bands after moistening the stone wall had positive noticeable effects on the accuracy assessment, 
even higher than expected. The inclusion of those three bands (red, green and blue) was fundamental to reach accuracies around 90 
percent. Unfortunately, it was not possible to verify the moistening effect by means of the near-infrared radiation; photographs were 
not available. 
6. CONCLUSIONS 
The supervised classification works effectively when it is applied to the identification and recognition of different kind of stones and 
pigments. This first study showed that the multiband classification is an optimal tool for the recognition and extraction of rupestrian 
paintings prior to restoration tasks. The effective delineation and shape recovery of primitive figures in rock art should help restorers 
to carry out their interventions, mainly when pigments are hardly visible and covered by pollution. 
The methodology developed in this study with multiband images seems promising to document opened-caves as well as to recover 
original paintings. Nevertheless, the more number of spectral and temporal bands, the more promising results. Further research needs 
to be done in order to know if this methodology is optimal to distinguish additional features or just inner materials. 
ACKNOWLEDGEMENTS 
The author would like to thank Dr. Martinez, chief of the Valltorta Museum, for his support and delivery of photographic material. 
This study was partly supported by the Spanish ‘Ministerio de Educación y Cultura ’ under grant PB98-1507-C02-02. 
REFERENCES 
Arte Rupestre del Arco Mediterráneo de la Península Ibérica, 1999. Conselleria de Cultura, Educació i Ciéncia-Generalitat 
Valenciana. 
Lerma, J.L. , 2001. Multiband versus Multispectral Supervised Classification of Architectural Images. Photogrammetric Record, 
17(97): 89-101. 
Puertas, Africa Adela, 2000. Procesado Digital de Imágenes de Arte Rupestre. Escuela Técnica Superior de Ingeniería Geodésica, 
Cartográfica y Topográfica, Universidad Politécnica de Valencia. Proyecto final de carrera. 
Vicent, J.M., Montero, I., Rodríguez, A.L, Martínez, M.I., Chapa, T., 1996. Aplicación de la Imagen Multiespectral al Estudio y 
Conservación del Arte Rupestre Postpaleolítico. Trabajos de Prehistoria: 53(2): 19-35. 
Perdrizet, F., Grussenmeyer, P., 1997. Multi-Scales Photogrammetry for Archaeology. International Archives of Photogrammetry 
and Remote Sensing. Vol. XXXII Part 5C1B, pp. 101-107.