1 Introduction
1.1 Related work
2 Materials and methods
2.1 Data set
2.2 Image features for biface search
2.3 Image-based search algorithm
2.4 Evaluation of image-based search algorithm
3 Biface classification contextual inquiry study
3.1 Participants
3.2 Materials
3.3 Procedure
3.4 Results
-
Blade versus flake The artefact will be classified as a blade or a flake. Blades are twice as long as they are wide; flakes are less than twice as long as they are wide.
-
Worked versus unworked A distinction is made between the cortex and core of the object. The cortex is the surface of the stone which has been worked by geological processes; the core is the interior part and is desired for tool working. As civilization is advanced, manufacture moved to using more of the core, while older artefacts exhibit more cortex. Artefacts may be distinguished by their type of removal: primary (removed from nearly unworked stone with substantial cortex remaining), secondary (a biface with only a thin seam of cortex created from stone closer to the core), or tertiary (flakes that have been created from the deep interior of the stone after primary and secondary strikes have removed the cortex and therefore exhibit a smooth, glassy quality).
-
Retouched versus non-retouched Once a removal has been struck from the core, the piece will either be left as rubbish (because it is small or brittle), picked up and used as is, or retouched. Retouched pieces are worked into a specific shape for a specific task, e.g. knapping to give it a sharp edge. The manner of retouching was reported to be of significant interest to archaeological research.
-
Colour The colour of stone bifaces from the same region may or may not be of a consistent colour and therefore may or not be diagnostic. Whether colour is of diagnostic use may depend on the data set and geographic regions under consideration.
-
Morphology Morphology refers to the shape of the biface, or the number, shape, and length of individual tines on the artefact. It is the most common method of matching finds to a geographic region or culture. Complications arise when some morphological differences are derived from other morphologies; for example, a tine could have broken off and retouched into another type of biface through knapping.
4 Biface image descriptors
4.1 Image preprocessing
4.2 Morphological features
4.2.1 Geometric features
Feature | Description |
---|---|
\(g_1\)
| Scale-normalized length of the artefact’s bounding box |
\(g_2\)
| Scale-normalized width of the artefact’s bounding box |
\(g_3\)
| Scale-normalized area (number of pixels in the artefact’s segmentation) |
\(g_4\)
| The scale-normalized breadth of the artefact at 20% of the distance along the length of the bounding box |
\(g_5\)
| The scale-normalized breadth of the artefact at 80% of the distance along the length of the bounding box |
\(g_6\)
| The ratio \(g_2/g_1\)
|
\(g_7\)
| The ratio \(g_4/g_5\)
|
4.2.2 Shape features
4.3 Texture descriptors
4.3.1 Uniform local binary patterns (ULBP)
4.3.2 Orthogonal combination of linear binary patterns (OCLBP)
4.3.3 Segmentation-based fractal texture analysis (SFTA)
4.3.4 Global phase congruency histogram (GPCH)
4.3.5 Angular radial phase congruency histogram (ARPCH)
4.3.6 Orientation-based phase congruency histograms (OPCH)
4.3.7 Gabor wavelet features (GWF)
4.3.8 Log-Gabor wavelet features (LGWF)
4.3.9 Binary texton features (BTF)
4.3.10 Fusion of LGWF and ARPCH (L–A)
4.3.11 Fusion of LGWF and BTF (L–B)
4.3.12 Fusion of LGWF, ARPCH, and BTF (L–A–B)
4.3.13 Fusion of LGWF, ARPCH, and SFTA (L–A–S)
5 Image-based search algorithm for bifaces
5.1 Feature extraction
5.2 Dissimilarity measures
5.3 Retrieval algorithm
6 Validation of the image-based search algorithm
6.1 Validation methodology
6.2 Individual metadata match performance
×
|
6.3 Normalized accuracy
6.4 Pairwise symmetric difference comparisons
×
|
6.4.1 Average number of disagreements
6.4.2 Differences in metadata matches for symmetric differences
×
|
6.4.3 Dominance
×
|