Home | Biodata | Biography | Photo Gallery | Publications | Tributes
[Back to Epigraphy List]

Epigraphy


Computer techniques of image enhancement in the study of a
Pallava grantha inscription

 Studies in Indian Epigraphy, Journal of the Epigraphical Society of India, Volume 2, 1975, 55-58 
also STAT 22/1975, September 1975

Gift Siromoney

THERE HAVE BEEN REMARKABLE developments during the last fifteen years in many areas of picture processing by computers1. Of special interest to epigraphists are the areas of image enhancement and line detection2. We present here the results of some experiments conducted on a fragment of a Sanskrit inscription 3 written in the Grantha script of the nail-headed variety. The inscription is from the Kaila$sana$tha temple complex at Kanchipuram erected by Ra$jasimha Pallava in the eighth century A.D. and is known as the Ran^gapata$ka$ inscription4 since it refers to a queen called Ran^gapata$ka$ .

Estampages of Ran^gapata$ka$ inscriptions were chosen for study. A fragment which reads : 
rjjitya garvvarn
= i$va Pushkara  de$vata$ya$h
was divided into two equal horizontal strips. Another strip representing a different estampage of the latter half of the fragment was placed on top and the three lines were photographed. The three lines read as follows:
ra de$vata$ya$
rjjitya garvvam = i$va Pushka-
ra de$vata$ya
The first and the last lines exhibit the variation between two estampages of the same fragment. The photograph is "read in" by the scanner unit of the computer and stored in a tape. One can call back the picture from the tape to be displayed on a television-like screen of the computer. The picture on the screen can be photographed by a Polaroid camera fitted to the unit. Figure 1 gives a Polaroid picture of the three lines used in our experiments and the eye-copy of the fragments is given in figure 2. We shall treat the inscriptions in figure 1 as "original" pictures.

The technique used for storing a picture in a computer tape can be described as follows. A picture is subdivided by the computer into tiny points arranged in rows and columns. Each point in a photograph is either black, or white or has a certain intermediate level of grey. If a part of a picture is absolutely white then the grey level at that point is reckoned as zero. If a point is taken from the darkest patch, the grey level will have a high value, depending upon the number of levels of grey we wish to recognize in the picture. The larger the number of grey levels the finer the details. Each grey level is measured by the scanner arid recorded as a number in the tape.

One of the interesting techniques of image enhancement is thresholding. Imagine a black and white picture having thirty two grey levels. Let us suppose that eight is fixed as the threshold value. (There are techniques for choosing the most efficient threshold). If any point has a grey level above eight it will be changed to black by the computer. Any point with a grey level up to eight will be changed to white. When the method of thresholding is used a ''black and white" picture with different intermediate shades of grey will be converted to a high contrast picture which has only white and black without any intermediate shades of grey. This picture is also called a binary picture since it has only two levels. One may try different levels of thresholding and choose that level which is most useful for the particular estampage. By setting the threshold high in our picture we will be ignoring finer details which might be of value.

Another simple technique used by computer scientists is complementation. By this method the negative of a given picture is produced by the computer. When a grey level picture of an estampage has been transformed into a two level black and white picture, complementation produces a high contrast picture with black writing on a white background. We have given here a number of examples of such pictures where the writing is shown in black. In the original the writing is in white on a black background produced by inking. Even if a photograph of an inscription is used instead of the photograph of an estampage the methods of thresholding and complementation may prove quite useful. When a computer is used the results can be displayed on the screen within a very short time. 

Averaging is another technique used in picture processing. The grey level value of each point is replaced by the average grey level of its neighbourhood. This is done quite easily by the computer. By this method isolated points will be smoothed out [Figure 3 (a)]. If thresholding is used after averaging, isolated points will get removed. This method is quite useful in many areas of picture processing but is not very effective in our experiment [Figure 3(b)]. We have also tried out the technique known as sampling where only one in ten points was sampled and the rest of the picture reconstructed by the computer on the basis of the sampled points [Figure 3(c)].

Since we are primarily interested in lines in the inscriptions, experiments using line detecting techniques were carried out. Pallava Grantha inscriptions do have "dots" to denote the "m" and the "h" sounds but the dots do not often stand out in estampages. Making use of the redundancy in Sanskrit it is possible to read an estampage with a high degree of success even in the absence of dots.

The line detecting technique used in our experiments was developed by Vander Brug and is called the semilinear line detector. It looks for a sequence of adjacent points along the direction of the line whose average intensity is darker than the average intensities of each of the adjacent sequences in the direction across the line. Semilinear detectors are not easily distracted by adjacent noise points and they tend to bridge small gaps in the line. Line detector tends to produce pictures with lines of uniform thickness even when the thickness of lines in the original picture is not uniform. If the line detector is used after thresholding we get very good results for estampages which are not "noisy" [Figure 4 (a)]. If the original is not clear but contains too many spots and smudges (in other words, noisy) then the line detector produces lines which are not in the original [Figure 4 (b)]. Threshold selection is a critical decision and we have given examples for different thresholds [Figure 4 (c) ]. Other line detecting techniques known as linear and non-linear detectors, if tried, might give similar results.

The techniques described here are quite general in nature and are applicable to different kinds of scripts and inscriptions found on stone, copper plate and other surfaces. It is hoped that, in future, computer methods of image enhancement will prove to be of value to those working with new inscriptions.

Acknowledgement:

The work described in the paper was carried out at the Picture Processing Laboratory of the University of Maryland, College Park. The author wishes to thank Professor Azriel Rosenfeld for his assistance in carrying out the work and Mr. Gordon J. Vander Brug for implementing the work on the computer. The support of the Homi Bhabha Fellowships Council is gratefully acknowledged.

Notes :

1. Azriel Rosenfeld, Picture Processing by Computer, New York, 1969.
2. Gordon J. Vander Brug, "Semilinear line detectors", Computer Graphics and Image Processing (forthcoming).
3. South Indian Inscriptions, Vol. I, pp. 23-24.
4. Michael Lockwood, Gift Siromoney and P. Dayanandan, Mahabalipuram Studies, Madras, 1974, p.59. In this inscription the chief queen (unnamed) of Narasimha II (or Ra$jasimha) is compared to Pushkarade$vata$ or Lakshmi. Ran^gapata$ka$, in our opinion, was the queen-mother the surviving wife of Parame$shvara. Ran^gapata$ka$ is compared to Pa$rvati. This view is different from the traditional view that$ Ran^gapata$ka$ was Ra$jasimha's queen.

Go to the top of the page


Home | Biodata | Biography | Photo Gallery | Publications | Tributes