JWA V1N2 - A DIGITAL IMAGE APPROACH TO ANALYSIS OF STONE TOOLS by Kell and Kellner

A DIGITAL IMAGE APPROACH TO ANALYSIS OF STONE TOOLS

Robert E. Kell and Grace Kellner

(Robert E. Kell ao172@freenet.buffalo.edu)

We are currently applying digitized computer techniques in studying the surfaces of stone grinding tools from Tell Ifshar. We are hoping to find clues about their history and use. This topic was suggested by Dr. Sam Paley. The objectives of this program are the following:

To replicate the surface shape and overall contour of these specimens for study in detail at sites far removed from their present location, and to make replicates of these specimens for possible destructive testing and use by other investigators.
To translate the surface shape and contour into images; to search these images for surface irregularities (wear marks, pits and fissures, and evidences of man-made features); to measure the size and shape of these irregularities; and to determine their average linear separation and density per unit area.
To generate these digitized images by use of an electronic camera and associated apparatus which produces computer files for storage and subsequent mathematical analysis.
To apply digital image processing methods, to locate and identify surface irregularities; to classify them into groups for further study; and to uncover information beyond that available through direct visual inspection. Processing methods now being considered include correlation, gradient and higher derivative calculation, and upper and lower bound thresholding.

These tasks may be carried out in three phases. The present discussion is a report on work currently in progress.

Phase One is the replication of the specimen surfaces into molds by using plastic impression material. This technique has proven successful for replicating very fine-grained features of various objects (e.g., teeth and human skin). Its use in the current context appears appropriate. This judgement will be checked by trials against similar objects (which need not be well- documented archaeological specimens).
Phase Two consists of using the molds prepared in Phase One to cast replicas of the original specimens. Dental stone (a very fine-grained, hard plaster) will be used for this purpose . It is available in several grades and colors. Plastic replication molds that have been prepared in phase 1 may be used at the preparation site, but they may also be transported to any other location where detailed studies of surface shape may be more conveniently carried out.
Phase Three requires an electronic camera and digitizing equipment to generate digital computer files suitable for mathematical processing. To obtain images useful for our purpose, the camera-digitizer combination must meet the following conditions:
1. The image must be stable and reproducible, so that quantitative measurements can be made and can be relied upon.
2. The electro-optical transfer function of each pixel must be a smooth function of the local image intensity, and the electro- optical response of neighboring individual pixels must be as similar as possible, to minimize data artifacts introduced by the photo-recording and file conversion process. Either linear or logarithmic response is acceptable.
3. The overall dynamic range of camera and file conversion should allow at least 8 meaningful data bits (256 gray levels) in the image files.
4. Geometric resolution (on any two orthogonal axes) should not be less than 0.005", and a much higher resolution such as 0.001" is desirable.
5. The transformation scale along the two principal axes should be the same.

Phases One and Two have already been carried out. Plastic replication molds have been prepared by Mrs. Kellner from a large number of specimens of grinding stones in Israel, at Tell Ifshar. These molds have been brought to Buffalo, where they have been used to cast plaster replicas. The replicas are expected to duplicate the original surface contours, including any significant irregularities. They do not carry the coloring of the original models, however; they are of a uniform color characteristic of the particular material chosen for casting.

This is an advantage in the imaging phase of the studies, since intensity of each image pixel will be determined by surface geometry unaffected by the varying color of the original specimen.

By carrying only the plastic replication molds from their site of origin to the places where facilities for making detailed studies exist, considerable savings have been made in the weight and bulk of the items which must be transported. Additional benefits of this approach are that extra copies of each specimen can be made for use in destructive tests, and that copies can be shared with other investigators.

Phase Three is currently under way, using a relatively inexpensive electronic camera. After reviewing the types available on the market, we chose the SONY Mavica-I, and have a ComputerEyes digitizer which works with a Macintosh computer to produces digital files. These choices were made on the basis of cost and easy availability, and to provide a means of gaining familiarity with this technique. The camera accepts a 2"X2"X0.25" minifloppy disc, carries its own (flash) light source, and allows 50 pictures to be recorded on each disc. Unwanted pictures can be erased and the space used again.

It was expected that the performance of this camera-digitizer combination would fall somewhat short of meeting the requirements outlined above, so a program was planned to evaluate its performance on the critical areas of resolution and image stability.

To test the resolution of the camera, a test target having a series of parallel lines was prepared and printed on an Epson LQ-570 dot-matrix printer. This printer allows printing at 360 dots per inch along one axis, and up to 180 dpi along the orthogonal axis. Several line spacing groups were included in this test target, with each group having a different spacing. Files for quantitative evaluation were then prepared which represent recorded image intensity as a function of position (x-y location) in the image. Examination of these files for pixel intensity along chosen image axes then gave a quantitative measure of the camera performance in discerning a single dark line, or in resolving a closely-spaced series of lines. It also allowed evaluation of the stability and uniformity of response.

It was immediately apparent upon examining these records that the test targets prepared with the Epson printer were inadequate to provide a valid test case. The lines drawn by the dot-matrix printer were neither sufficiently straight nor uniform in weight to provide a valid test. The simple target design was also inadequate to discriminate against intensity variations caused by overlapping of pixel response to the low intensity skirts of the camera lens point spread function, several resolution elements away from its central axis. The camera performance is still undetermined, but it appears good enough to justify continuing with this approach.

We are now experimenting with improved target designs and better printing technology, in an attempt to get a valid judgement concerning the resolution and stability of the camera.