Cultural Heritage Applications: Archaeological Ceramics and Building Restoration - On Statistical Pattern Recognition in Independent Component Analysis Mixture Modelling - page 128

wall of a restored dome of a Basilica. ICA is used as pre-processor to obtain
enhanced power signal B-Scans of the wall.
6.1 Chronological Classification of Archaeological Ceramics
6.1.1 Introduction
Determining the historical period of archaeological ceramic shards is important for
many archaeological applications, particularly to reconstruct human activities of
the past. In fact, the standardization of an efficient and non-destructive
testing(NDT) method for ceramic characterization could become an important
contribution for archaeologists. Chemical, thermo-luminescence, and other anal-
yses have shown to measure the age of ceramics accurately, but they are expen-
sive, time-consuming and involve some destruction of the analyzed pieces [ 2 ].
Relative dating by comparison with ceramic collections is non-destructive but very
inaccurate [ 2 ].
Ultrasound has been used in archaeological applications such as ocean explo-
ration to detect wrecks, imaging of archaeological sites, and cleaning archaeo-
logical objects [ 3 - 5 ]. In this application, we consider a method to sort
archaeological ceramic shards based on ultrasonic NDT. This method aims to be
economic, fast, precise, and innocuous for the ceramic pieces. It consists of three
steps: measuring by the through-transmission technique, extracting features from
the measured ultrasonic signals, and classifying the feature set in classes corre-
sponding to historic or protohistoric periods.
The estimation of the chronological period of an archaeological fragment is not
a straightforward work, especially if we consider that the fragment might be
moved from its context of origin due to migrations, wars, or trade exchange, etc. In
addition, some external features used for classification of archaeological objects,
such as particular shapes and decorations, might be not evident in the fragments,
and thus these aspects would not provide information for a correct classification of
the fragments.
Through-transmission was selected because the ceramic produces large
attenuation to the propagation of ultrasound, so the pulse-echo technique cannot be
implemented at the required operating frequency. Time, frequency, and statistical
features (to be described later) were extracted using standard signal processing
techniques. The characteristics of the classification problem offer a good case
study for testing advanced classifiers, like those based on modelling the underlying
statistical densities of the feature space as mixtures of independent component
analyzers(ICA).
In consequence, we dedicate Sect. 6.1.2 to presenting the ultrasound through-
transmission model from a linear system perspective and to defining the selected
features. Then, in Sect. 6.1.3 we present the rationale for these classifiers and