Raster
In 1957 a drum scanner at the National Bureau of Standards cut a photograph into squares a quarter of a millimetre across and handed SEAC one binary digit for each of them: dark or light, and nothing else. The picture was 44 millimetres square, which makes 176 squares along each side and 30,976 digits in all. The image everybody has seen from that work looks like a grey photograph. No square in it holds a grey. The reference the scanner compared against was varied while it scanned, so the share of black squares follows the darkness of what was underneath, and the eye assembles a shade out of a density of dots. This page draws its own field, puts it through both mechanisms, and measures the second one against the sentence the paper makes about it.
A picture is a grid of numbers, and the first one had one bit in each square
A picture becomes a grid of numbers by being cut into squares and each square given a value. In 1957 that value was one bit. The scanner compared the light off each square against a reference set by hand on a potentiometer, and fed SEAC a 1 for dark and a 0 for light. Move the reference and the whole picture moves with it, because there is one setting for the entire scan and nothing in between black and white.
| quantity | value | where it comes from |
|---|
The image everyone has seen from this work looks like a grey photograph, and no square in it holds a grey. The paper says how: the reference was varied while the scan ran, so that a square darker than the reference at that instant came out black and a lighter one did not. Over an area the share of black squares follows the darkness, and the eye does the rest. The greys are a density of dots, not values, and the panel below measures that rather than asserting it.
| greyness of the source | share of squares black | difference |
|---|
These ran in this browser when the page loaded. Each claim, whether it held, and the number behind it.
| claim | held | measured |
|---|---|---|
| 176 by 176 is 30,976 squares, which is the figure the paper prints | yes | 176 x 176 = 30,976 |
| and 44 mm across 176 of them makes each one 0.25 mm, which the paper also prints | yes | 44 mm / 176 = 0.25 mm a square |
| and at 44 binary digits to a SEAC word that is 704 words, which is the figure the paper prints | yes | 30,976 / 44 divides exactly and lands on the number Kirsch states, so this is a round trip against the paper rather than a figure taken from it |
| at one fixed threshold a flat grey comes out all black or all white, never grey | yes | a grey of 0.5 gives a density of 1 and a grey of 0.4 gives 0: two levels exist and neither is a shade |
| sweeping the threshold makes the density of black squares track the greyness, to within 0.000 | yes | measured over 17 grey levels at a period of 16 squares; the paper says the density of uniformly black spots is proportional to the blackness of the original |
| and a darker input never produces fewer black squares | yes | the density is non-decreasing across every step of the sweep |
What is real here, and what is not
This is not Kirsch's photograph, and it does not pretend to be
The field on this page is drawn by this page: a wedge with a disc in it, chosen so a reader can see a gradient and an edge at once. The famous 176 by 176 image is a portrait of Kirsch's son and NIST holds it. Reproducing it here would be a claim about a scan that happened in 1957 with equipment this page does not model, and the mechanism is what is being shown.
The arithmetic reproduces the paper's own figure, and it was once said not to
Three figures, and the paper prints all three. The picture is 44 mm by 44 mm and is digitalized into 176 by 176 or 30,976 binary digits; the scanner's output is arranged in SEAC input words of multiples of 44 binary digits; and the entire picture with one binary digit per square occupies 704 words of SEAC memory. Computing 30,976 over 44 and landing on 704 is therefore a round trip against the document, which is worth more than a derivation of something it omitted. An earlier version of this page said the paper did NOT print 704 and presented the number as this page's own. That was wrong, and the way it was wrong is worth keeping: the archived extraction of the scan had silently dropped four of its nine pages, this sentence among them, and a search that found nothing was read as the paper containing nothing. The archive is the OCR now, and la-source-audit checks a text layer page by page rather than in total, because the five pages that did carry text covered for the four that did not.
Nothing here is in colour, and there was no grey either
There is no colour anywhere in this work. The scanner produced one bit a square and the roster's blurb says so; an earlier draft of that line said a colour is three numbers, which is true of later machines and not of this one. There were no grey levels either: the paper describes obtaining several levels of greyness by making several scans at different manual settings, which is a different mechanism again from the swept threshold shown here.
The sweep is a mechanism, not a reconstruction of the waveform
The paper gives the varying threshold as a waveform at approximately half the strobing frequency. The period here is the reader's, in squares, because the point is that density follows darkness and not that this reproduces the exact 1957 waveform. At the shortest periods the tracking is coarse, which is visible in the table rather than hidden. The archived scan is two columns and its text layer interleaves them, so this page quotes only the two fragments that survive contiguously and describes the rest.
The scanner is older than the paper, and the entry is dated to the paper
The paper records that the scanner was first connected to SEAC in November, 1956, and it was published in December 1957. The roster dates this entry to the document, which is the same choice Context Switch makes with Atlas and DMA with the 709, and the prose says the equipment ran the year before rather than smoothing it over.
Not the first raster scan, and not a claim about television
Scanning a picture in lines is decades older than this and belongs to television. What happened here is that the result was put into a computer's memory as numbers it could compute on, which is a different claim and the one this page makes.
Sources
- Russell A. Kirsch, Lawrence Cahn, Charles Ray and G. H. Urban, Experiments in Processing Pictorial Information with a Digital Computer, Proceedings of the Eastern Joint Computer Conference, December 1957, pages 221 to 229, from NIST. Page 222 for the scanner and its 0.25 mm squares, the SEAC word format and the November 1956 connection; page 223 for the 44 mm picture, the 176 by 176 grid and the halftone.
- Logical Art, the studio this belongs to.