Comparación entre el método basado en píxeles y el basado en objetos para el análisis de fachadas de edificios históricos

Abstract

 Techniques of digital image processing give the possibility to  detect damages, such as moisture or biological changes on  surfaces of monuments in a non-destructive way. Traditional  classification methods are all pixel-based and do not utilise  the spatial and context information of an object and its  surroundings, which have potential to further enhance digital  image classification. In this study, we compared the application of pixel-based  classification and the object-oriented classification technique  using multispectral photographs taken with a Fujifilm IS-Pro  digital single lens reflex camera in order to detect and locate  damages affecting biocalcarenite stone employed in the  construction of the Mosque of Cordoba (Andalusia, Spain).  The best results were achieved with object-oriented  classification obtaining an overall classification accuracy of  96.55% and an excellent kappa statistic (0.944). These results  have shown that object based feature extraction can result  in powerful solutions for the interpretation of high resolution  data. It is proved that object-based approach to classify  imagery is much better than the traditional classification  techniques, such as pixel-based.