In this paper, we apply supervised neural networks (Backprop. learning algorithm) to the classical problem of statistical hypothesis testing. Processing experimental use wear in lithics we have found some contraintuitive results using standard tests, which can be solved using the non-linear discriminant power of Neural Networks. Specifically when archaeological data do not fit parametric distributions, Supervised Learning algorithms appear as an alternative approach. Our particular case study is a set of digital images of experimental data showing use wear as a result of work actions. We have used replicated lithic tools in order to find similarities between use wear identified in experimental data. Previous studies shown that there is not an single discrimination rule to associate cause (kinematics) and effect (wear). DESCRIBING USE WEAR AS TEXTURE PATTERN Archaeologists studying lithic remains usually wish to determine whether or not these stones have been used as tools and how they were used. The best way to do this is through the analysis of macroand microscopic traces of wear generated by the use of the tool (Fig. 1). ORIGINAL UNALTERED SURFACE ALTERED SURFACE AFTER HUMAN WORK Fig. 1. Distinguishing altered from unaltered micro-surfaces. The main assumption is that the surface of artefacts have specific features because of the way they have been made, or the way they have been used. Tools are made of solid materials and have rigid bodies which resist stress. As any other physical entity, objects have surfaces, which can be defined in terms of their size, shape, composition and location. Texture can be defined then as the pattern of variability within this surface of those basic properties (Pijoan et al. 1999, Barceló et al. 2001, Adán et al. 2003). In the case of tools, given that use and production make important alterations in surface features, we can use texture information to understand how the object was made and/or used (human work) (Fig. 2). Texture variations due to human work are evident, and vary according to the following causal factors: • Movement: longitudinal (cut), transversal (scrape),... • Worked Material: (wood, bone, shell, fur, etc.) the effects of its physical properties (hardness, wetness, porosity, plasticity, etc.) on the tool activity surface Fig. 2. Texture differences between lithic tools used in different ways. A: original andesite texture before using; B, Result of the alteration in surface A when the tool was used scrapping fur. C, A different raw material (obsidian) with texture features produced through wood scrapping. We usually represent textures using images. What we are looking in that image is the patterning of luminance values across all pixels. Images have texture (luminance variation), which can be used to represent the variation of the object surface properties (surface texture). The texture of different images should allow us to discriminate between image groups with some characteristic pattern of luminance variation (Adán et al. 2003). Texture is then described as the relationships of luminance values in one pixel with luminance values in neighbouring pixels (Pijoan et al. 1999, Barceló et al. 2001, Russ 1995, Fontoura and Marcondes 2001). These values can be modelled as forming a set of regions, consisting in many small subregions, each with a rather uniform set of luminance values. In our case, these values are defined as grey levels. A group of related pixels can be considered as a texture minimal unit, sometimes called texel – texture elementTexture patterning in an image should be described as associations between texels . We define luminance discontinuities (region in an image) as texels, if a set of local statistics or other local properties of the average density function are constant, slowly varying, or approximately periodic. Our goal is to segment those texture elements, in order to be able to study their variability in shape and spatial location (Fig. 3). Texels may be geometrically described and measured or they can be “identified” subjectively in the microscope image as texture primitives; the researcher “sees” stries, polished areas, scars, particles, undifferentiated background. Even the “intensity” of a trace has also been determined subjectively, introducing attributes like “poor”, “high”, “developed”, “greasy”, etc. However, we should calculate their formal and relational properties, using their variables of shape, size, composition, and location.
Download Full PDF Version (Non-Commercial Use)