Computerized Plating of the Large Hermes Heads of Greece

by Louis Basel

(This article was originally published in The American Philatelist in the May, 1985 issue. It was awarded the Apfelbaum prize for the best article in that journal for the year 1985. It also won a silver medal at the Hellas 84 Exhibition in Athens in 1984, where it was first exhibited. Computerized plating of the Large Hermes Head stamps was probably one of the first applications of the personal computer to technical philatelic research. Although this original computer program was abandoned after a few years of its successful use, it was replaced by a newer program using a faster programming language. Unfortunately, the supplier of the digitizing hardware and software used to convert the analog video image to digital format went out of business and it was impossible to convert these programs for use in modern Windows-based computers. Recently, a new program has been written using the CorelDraw and Excel programs with scanned images of the stamps. Another more automatic program using the C++ programming language is currently under development.)

Introduction

The large Hermes heads of Greece, the first stamps to be issued by that country, were issued on October 1, 1861. Albert Barre engraved the master die in Paris, fabricated the plates, and arranged to have E. Meyer print a quantity of stamps to be sent to Athens.

All subsequent issues up to 1875 were printed in Athens by the Greek government post office. However, at that time, new denominations were required, and Barre was again engaged to provide new plates and a quantity of stamps.

Later, additional printings were made in Athens from these new plates, as well as from the original plates. All of the seven original plates except the 80-lepta were used for approximately twenty-five years (Figure 1). The large Hermes heads were withdrawn from circulation in 1891.¹

1. T. Constantinides, A Study of the Stamps of Greece (Athens, The Hellenic Philotelic Society, 1933).

Plating of the large heads is facilitated by the many plate varieties, a few of which were described by Beckton² in 1896, and by Pemberton,^3,4,5 Morris,⁶ and Brunel⁷ during the next thirteen years. Plate varieties continued to be reported and, in 1928, the detailed study of Munk and Argyropoulos⁸ was published in “Kohl's Handbook”, with a list of the major varieties and their plate positions.

2. W. Dorning Beckton, “The Stamps of Greece”, The Philatelic Record, Vol. XIX (April-Aug. 1897), pp. 91 et seq.
3. P.L. Pemberton, “Notes on Greek Stamps,” Philatelic Journal of Great Britain, Vol. XI and XII (Feb. 1901 - Feb. 1902), pp. 20 et seq.
4. P.L. Pemberton, “Descriptive Catalogue of European Postage Stamps - Greece;” Philatelic Journal of Great Britain, Vol. XIII-XIV (Nov. 1903 - Jan. 1904), pp. 185 et seq.
5. P.L. Pemberton, “The Stamps of Greece”, The Philatelic Journal of Great Britain, Vol. XXI and XXII (Jan. 1911 - Sept. 1912), pp. 6 et seq.
6. T.W. Morris, “The Stamps of Greece”, The Stamp Collector, Vol. X and XI (July 1906-Aug. 1907), pp. 95 et seq.
7. Georges Brunel, “Les Emissions des Timbres Grecs”, Philatelic Journal of Great Britain, Vol. XIX (July 1909), pp. 144-45. Original published by Charles Mendel, Paris, 1909.
8. H. Munk, "Kohl's Handbook: Greece Parts I-V," trans. H.G. Zervas, The Collectors Club Phitatetist, Sect. II (April 1943 - July 1950) (original1y published in Germany in 1929).

In 1929, Groom⁹ published his first work on ink spots which were constant during several printings, and he showed how they could be used to identify plate position. In 1933, the Hellenic Philotelic Society published the comprehensive “Etude” of Constantinides¹ listing all of the then known varieties.

9. T. Groom, “The Crust-Flaws of Stamps of the First Type and Their Uses”, The London Philatelist, Vol. XXXVIII (1929), pp. 109 et seq.

The work of Garas¹⁰ on varieties of the plates was published in 1955, and that of Alfieris¹¹ on varieties of the 20-lepta of 1868 in 1956. The most detailed study of plate varieties and ink spots was published in 1979 by Bellas,¹² who gave definitive descriptions of each plate position of the 20-lepta.

10. N. Garas, Greece - The Large Head of Mercury, The Varieties of the Plates of the Postage Stamps, 1861-1882 (Athens, published by the author, 1955).
11. N.S. Alfieris, “The First Type of Greece (1861-1884): The 20 Lepta of 1868”, The London Philatelist, Vol. LXV (February 1956), pp. 30-44.
12. U. Bellas, The Head of Mercury - Generalities & Reconstitution of the 20 Lepta Plate (Paris, published by the author, 1979).

These and other studies indicate the great attention given to the varieties and plating of the large head stamps. In spite of all this work, only the 20 1epta have been described in as detailed and systematic a manner as Bellas. However, even with the aid of his study, the determination of plate position for the early 20-lepta issues is very difficult and time-consuming because they lack the many unique and often prominent markings which appear after the plates had been in use for a number of years.

The purpose of this study was the development of an automated technique to facilitate plating those large Hermes head denominations which have not yet been completely described in the literature. It was hoped the method would be capable of determining quickly the plate position of even early issues devoid of prominent markings. The procedure was based on the 20-lepta stamps because they have been so well described by Bellas, and was then tested on a number of 20 lepta stamps whose plate position previously had not been determined. The test equipment, the procedure used, and the results of this investigation are described below.

Fabrication of the Plates and Their Varieties

It is necessary to present some details on the fabrication of the plates for a better understanding of the plating method described below. Each of the seven original plates had 150 bronze clichés arranged in fifteen horizontal rows of ten clichés each. Barre, the fabricator of the large head plates, had to manufacture 150 clichés for each of the seven different denominations, which amounted to a total of 1,050 clichés. Rather than engraving seven master dies, one for each denomination, he engraved only one die, leaving the numerals blank. After the 1,050 clichés were fabricated, hand punches were used to manually strike the desired numerals onto the two blank spots of each cliché. Because these numerals were struck one at a time, their position on the clichés varies both horizontally and vertically, and in some instances the numerals are slanted (Figure 2).

Bellas used a visual method for determining the degree of numeral shift or its lack thereof, and was, therefore, able to indicate only major, minor, or zero shift. This degree of differentiation is inadequate for the precise determination of plate position using numeral positions alone because so many of the plate positions have similar shifts in the horizontal and vertical directions.

Another marked difference between the clichés is the spacing of the outer frame tine from the internal portion of the engraving. Many specialists have studied this variation using the French terms “filet distant” (wide spacing) and “filet serré” (narrow spacing) (Figure 3); the plate positions where each occurs have been identified.

Bellas has observed that on some stamps this spacing varies, being wide at one end and narrow at the other, while on other stamps it varies in more complicated ways. On a few stamps, the frame line spacing is narrow on three sides and normal on the other side of the same stamp. However, because of over- and under-inking, stamps from the same plate position often display different frame line spacing, and it is difficult to use this measurement as a reliable means of identifying plate position.

In a similar manner, inner white spaces separating the internal colored portions of the engraving vary in distinctive ways from stamp to stamp, in some cases showing an abrupt widening, usually at one end or the other (Figure 4).

These three varieties are true plate varieties in that they appear on the first stamps printed in Paris and on all later issues of stamps from the same positions. Bellas suggests that the clichés were retouched after they were coined from the master die, causing these variations in frame line and internal line spacing.

In addition to these plate varieties, there exist many others which are constant for a particular plate position and for a period of time, but which are not observed on al1 issues. These were generally caused by damage to the plates or by inadequate cleaning. Examples of plate damage include scratches, shortened or distorted shading lines, dented or broken frame lines, small white spots in the normally colored portion of the cliché, etc.

Inadequate cleaning resulted in the appearance of clumps of dried ink on the cliché which were then printed as small colored spot on the stamps. These ink spots were constant on stamps of the same plate position, and often occurred over a number of years until the plates were cleaned. In some cases, a partial cleaning removed some of the ink clumps but not all; in others, new ink spots appeared after a previous cleaning and are found together with some of the original ink spots which had not been removed. Stamps printed during 1868 and 1869 have a very clean and neat appearance, indicating that a major cleaning had taken place. These stamps are appropriately labeled the “Cleaned Plate Issue.”

Examples of a few of the “temporary” varieties are shown in Figures 5 to 8. Note that some of these temporary varieties may indeed be true plate varieties, but insufficient examples of Paris issue stamps were available to verify this point. Figure 1, position 10, also shows a white spot behind the neck of the Hermes figure, a variety which, to my knowledge, has not been reported to date.

    

Another important variety is the heavy frame line of sheet margin stamps on the side (or two sides in the case of corner stamps) adjoining the margin of the sheet, Bellas states that this variety appears only on stamps printed in Athens using a felt underlay beneath the sheet of paper on which the stamps were being printed. The Paris issue, the first Athens printing, and the German Workers Issue of 1870 did not use the felt underlay, but instead used the “à sec” method in which a découpage made of several layers of paper was placed under the sheet to be printed. (See Constantinides’ “Etude” for a description of both techniques.) Figure 2 shows the heavy frame line of Pos. 144 on the bottom side of the stamp.

Description of the Plating Method

The plate positions of most large head 20-lepta stamps can be determined easily using the Bellas study as a guide. However, the early issues were devoid of most of the temporary varieties, and plating of these early issues must. be based on the constant varieties. This is often difficult, because in previous studies the constant varieties were defined in a very qualitative manner. Of the 150 positions on the sheet, Bellas identifies twenty with wide frame line spacing and twenty with narrow spacing, and he describes the remaining positions as normal spacing.

Similar descriptions are used to define the location of the numerals "20" in the lower inscription block. Because of this rather imprecise differentiation, and because so many positions have similar constant varieties, the early large head issues and even some of the later ones, made after a cleaning of the plate, are extremely difficult to identify.

The 20-lepta large heads were used as the basis for this study because of the availability of the Bellas work on the 20 lepta, and because plate positions of the other six denominations have never been described in such detail.

It was decided that one or more of the constant varieties should be used as the means of determining plate position because it would be effective in plating early issue stamps. The location of the numerals in the lower inscription block was selected because of the greater number of variables available - i.e., the horizontal and vertical shifts of the two sets of numerals on either side of the “ΛΕΠΤ” inscription.

In this study, a video camera with a macro lens was used to photograph the stamp and to display the enlarged image on a video monitor. An IBM Personal Computer with a Tecmar digitizer board was used to digitize the image and to store it in computer memory for later processing. Figures 9 and 10 show the equipment used; they are listed in Table 1 along with the approximate cost.

    

The digitizer converts the light intensity of each picture element (pixel) to a digital value in a scale of 0 to 255. (black = 0) white = 255) for each of 255 locations in the horizontal direction and 255 locations in the vertical direction. This gives a total of more than 64,000 pixels whose light values are stored in computer memory. However, in displaying the enlarged image of one set of numerals and adjoining dots, only a portion of the screen is used, which results in a recording of about 10,000 pixels.

The two white dots on either side of the numeral “20” were engraved on the master die and are therefore in fixed locations on all plate positions. These dots provide handy reference points for measurement of the horizontal and vertical deviations of the numerals. In this procedure, a pattern recognition technique was used to locate the X (horizontal) and Y (vertical) positions of the centers of the two dots. By simple arithmetic, the X and Y values of the median point between the two dots were calculated (Figure 11).

A similar pattern recognition technique was used to locate the center of the “20”. Then the median dot X was subtracted from the “20” center X to determine the X deviation. The calculation was repeated for the Y values to obtain the Y deviation. This procedure was carried out for both the 1eft and the right numerals, resulting in four values for each stamp.

A reference dot image was created in computer memory for use in locating the center of the dots. This image was first located in the upper left corner of the left dot scan area, and a comparison was made between the light and dark areas of the reference dot and those of the stamp's digitized image. The number of similar pixel values was counted and retained in memory.

The reference dot was then moved one pixel to the right, and another count of matched pixels was made. If the second count was greater than the first, the second count and its X and Y positions were retained, while the first were discarded. This procedure was repeated until the end of the horizontal scan line was reached, at which point the dot reference was moved to the extreme left and down one pixel to start the next scan. After each comparison, if the new count was greater than the previous maximum, the new match count and the corresponding X and Y positions were retained.

When the dot reference scan was completed over both the X and Y ranges of the selected scan area, the maximum match count and its X and Y positions were stored as the center of the stamp image dot.

A similar procedure was used for the “20” using a reference “20” created in memory. The shape of the reference “20” was based on several images of Paris issue and other early issue stamps that had clean, undistorted shapes. The X and Y positions of the right dot were then determined in the same manner, resulting in three X values and three Y values. The calculations described below were made by the computer to determine the X and Y deviations:

X1 = center X position of left dot
X2 = median X position between dots = (X1 + X4)/2
X3 = center X position of “20”
X4 = center X position of right dot
YI = center Y position of left dot
Y2 = median Y position between dots = (YI + Y4)/2
Y3 = center Y position of “20”
Y4 = center Y position of right dot DX = X deviation of “20” =X3 - X2
DY = Y deviation of “20” = Y3 - Y2

Before proceeding to test the method on stamps whose sheet positions were unknown, it was first necessary to determine standard X and Y deviations for each plate position. This was done by measuring the deviations for the left and right numerals of five stamps from each position (a total of 750 stamps). The positions of these stamps had been previously determined visually using the Bellas information.

In a few cases, deviations for one stamp of a given position were found to vary greatly from those of the other four stamps, indicating that the stamp's position had been incorrectly identified by visual means, and its deviation values were not used. The averages of the four remaining sets of deviations (X and Y values for the left and right numerals) were used as the standard deviations for that position.

After standard deviations for the 150 plate positions were measured and stored in a data file on a computer diskette, X and Y deviations were measured for several stamps whose plate positions were unknown. A computer program was used to compare the deviations for the unknown stamp with each of the 150 standard deviations to find the closest match. The “least squares method” was used in a computer program to calculate the best fit. The unknown stamp was then checked visually against previously identified stamps of the position selected by the computer. The computer choice was correct in most cases but, in a few cases, errors were made.

Figures 12 and 13 show statistical distribution curves for the X and Y deviations of the left and right numerals. Note that the numerals are clustered around the zero X and Y deviation points, and that, for a few plate positions, the deviations are so slight the computer is unable to distinguish between them. However, it is interesting to note the rather wide deviations of some numerals, ranging from - 9 to + 14.5 pixels for the X deviation and from - 5 to + 7 for the Y deviation.

    

Another situation that aggravates the problem is that, on many stamps, the dots and “20’s” are distorted by dried ink or by over-inking of the plates. Because the reference dot and “20” were created with undistorted shapes, the pattern recognition program did not always properly locate the X and Y positions of these distorted figures. An example of distorted numerals is shown in Figure 14.

To improve the method, it was decided to measure additional parameters that would be useful in differentiating between the various plate positions. During the course of the above measurements, it was observed that on many stamps the numerals were considerably slanted. Two plate positions with similar X and Y deviations might have numeral slants that were quite different, which could be used to distinguish them.

The procedure used was to place on the video monitor at a fixed position the computerized outline of a “20” which was perfectly horizontal. Then the stamp image was displayed on the monitor. The stamp was moved under the camera in the horizontal and vertical directions and then rotated until the stamp “20” was exactly matched to the reference outline.

A microscope stage positioner with fine controls was used to align the stamp properly. If the numerals were slanted on the stamp, the two dots would be displaced in the vertical direction, the degree of displacement indicating the degree of slant (Figure 11). The X and Y deviations were measured at the same time as the slant, and the difference in the Y positions of the two dots was used to calculate the slant. Figure 15 shows the statistical distributions for the slants of the left and right numerals; they exhibit typical bell-shaped curves. The range of slant values was from - 10 to + 10 pixels. The slant was calculated as follows:

Slant = Y4 - YI pixels
Slant = Arctan [C*(Y4 – Y1)/(X4 – X1)] degrees
C = correction factor for camera aspect ratio

Because the frame line spacing varied on many of the plate positions, it was decided to test it as a possible parameter for use in conjunction with the previously described parameters. However, it would have been too time-consuming to measure the frame line spacing at a sufficient number of points to define completely the many variations along each edge of many stamps as noted by Bellas. It was therefore decided to make this measurement at only one point, directly below the left dot of the left numerals. The procedure used was simple and took only about twenty seconds for each stamp.

Figure 16, a sketch of the lower-left portion of the stamp, shows the measurements made to determine the frame line spacing. Dimension A is the width of the white space between the frame line and the edge of the lower inscription block. This is the dimension which probably was noted by previous investigators using visual means. Its use yields inconsistent results because variations in the degree of inking produce significant differences in its width for stamps from the same position.

Dimension B is the distance between the center line of the frame line and the edge of the inscription block. Although this measurement was more consistent than dimension A, values for several stamps from the same position still varied too greatly to be useful as a distinguishing parameter.

It was then decided to measure the distance between the center line of the dot and the center line of the frame line (dimension C) because both center lines should be relatively independent of the degree of inking. Measurements of this dimension on five stamps from each of the 150 positions confirmed that consistent values could be obtained for stamps of the same position.

To make these very small measurements (0.89 to 1.09 mm), the digitizer program places on the image screen a point cursor which can be moved one pixel at a time in the horizontal or vertical directions by pressing the arrow keys on the computer keyboard. The X and Y positions of the cursor are displayed on the computer monitor.

The cursor was placed in the center of the frame line directly under the left dot of the left numeral, and the Y position was noted. The cursor was then moved to the center of the dot and its Y position was noted. The difference between these two Y values (dimension C) was the value used as one of the parameters in the plate position determination procedure. The filet spacing measured for each of the 150 plate positions varied from 40 to 49 pixels, with a statistical distribution as shown in Figure 17.

The last parameter used was the heavy frame line of margin stamps. Although it was realized that large Hermes head margin stamps which were printed using the “à sec” method do not always exhibit this heavy frame line, it was used because it is so distinctive when present, and because relatively few of the large heads were printed by the “à sec” method. This determination can easily be made by a visual examination. If present, it immediately limits the possible plate positions for that sample to thirteen each for the left and right margin stamps, and eight each for the top and bottom. The four corner-position stamps with two heavy frame lines are unique.

There are thus a total of seven computer measurements (two DX and two DY deviations, two numeral slants, and the frame line spacing) and one visual determination of the presence or absence of a heavy frame line. In making the measurements of five known stamps from each of 150 positions, it was observed that the variations from the mean value of some of the parameters for stamps of the same position were greater than others. A statistical calculation was made of the 750 samples measured to determine the standard deviation (sigma) from the mean values of each position.

If a standard deviation is assumed, the value of sigma for a particular parameter indicates that the deviations of sixty-five percent of the measurements of that parameter will be within plus or minus sigma deviation from the mean value. These sigma values were used as weighting factors for each of the parameters in making the least squares comparison of an unknown stamp with each of the standard parameters of the 150 plate positions. The calculated sigma values for each of the measured parameters are given below:

To illustrate the significance of these calculations, measurements of one stamp from position 127 are presented below; accuracy notations are based on these sigma values (rounded to the nearest 0.1). In the same table are presented indications of a heavy frame line, “0” denoting its absence and “1” its presence.

The pixel spacing can be converted easily to millimeter values by using conversion factors which were determined by placing under the camera a reticle with divisions of 0.1 mm. Because of the aspect ratio of the camera, the pixel spacing is different for the X and Y directions. The measured values were 39.6 pixels per mm in the horizontal direction and 44.8 pixels per mm in the vertical direction.

Test of the Plating Procedure

To test this computerized plating procedure, the parameters noted above were measured for 102 unknown stamps. These parameters were compared with the standard parameters of each of the 150 plate positions. In the least squares method, the value of each sample parameter is subtracted from the standard value of the same parameter; the difference is squared and multiplied by the appropriate weighting factor; and the sum of the weighted squared differences for the seven parameters (plus the presence or absence of a heavy frame line) is calculated. The plate position with the lowest sum represents the plate position selected by the computer for the unknown stamp. This procedure takes the computer about three minutes for each unknown stamp. To provide for possible errors due to distorted dots and “20’s”, the sum of differences squared was printed for the five closest positions. Results of this test are presented below:

In plating large numbers of stamps, it is important to consider the amount of time required to determine the plate position of each stamp. Although the visual method took only a few minutes for stamps with prominent varieties or ink spots, thirty minutes to more than one hour were required for stamps of the early issues and those printed from cleaned plates and, in many cases, the visual determination was in error.

Using the computerized method described, the time required to digitize images of each stamp and store them in memory took about three minutes. The pattern recognition program, which located the X and Y positions of the dots and “20”, the comparison of the sample’s parameters with the 150 standard parameters, and the printing of the five closest positions took about six minutes. This is a total of about nine minutes per stamp. The time required for a visual check of the selected position should also be included, but, as this requires a comparison of only one or two positions, it takes only about one minute. The total time per stamp was thus about ten minutes.

The computer operator's attention was required only during the first part of the procedure. Therefore, to minimize the operator's time, the digitized image and frame line information were stored on diskette for twenty-one stamps (forty-two images), requiring about one hour. The computer then automatically retrieved one image at a time from the diskette, processed it, and printed the results, taking about two hours for the twenty-one stamps without any operator attention. The visual check again required personal attention, for twenty-one stamps about twenty minutes of time. The total time for twenty-one stamps was thus about three and a half hours, less than half of which required personal attention.

Obviously, the computerized procedure not only saved considerable time. It also gave more accurate results than the visual method, especially on stamps without ink spots or other obvious markings.

Note that the procedure requires clear images of the dots. Stamps having one or more dots completely covered with a postmark could not be used in this procedure.

Because about thirty-five percent of several hundred stamps inspected had postmarks covering dots either partially or completely, a semi-automatic technique was used to determine the plate position of those stamps with partially covered dots. This method could also be used for those stamps whose figures were so badly distorted that the computerized technique gave erroneous results. In this procedure, the cursor was moved to the visually determined centers of the dots and the “20”, and the X and Y positions were noted. These values were then entered into the computer, a comparison with the 150 standard parameters was made, and the five closest positions were printed.

Conclusion

The computerized procedure for the plating of the large Hermes head stamps of Greece has proved effective in quickly and accurately determining plate positions of a large number of stamps, even those of early issues and others without any prominent distinctive markings. Its big advantage is in narrowing the number of positions against which the unknown stamp must be checked visually.

The computer made erroneous selections for only a small number of stamps. In those instances, a brief visual check easily determined that an incorrect choice had been made, and the stamp could be put aside for a semi-automatic determination later.

It is proposed that the procedure described above be used to plate other denominations of the large Hermes head stamps and to describe their plate varieties in the same manner as Bellas did. This procedure might also be applicable to the plating of issues of other countries which exhibit dimensional variations of the engraving.