55 
med media: 
99. pp. 17- 
ie National 
a profound 
Information 
mal Library 
I discussion 
RE-SEQUENCING A HISTORICAL PALM LEAF MANUSCRIPT WITH 
BOUNDARY-BASED SHAPE DESCRIPTORS 
D. Akca , A. Gruen 
Institute of Geodesy and Photogrammetry, Swiss Federal Institute of Technology (ETH) Zuerich 
ETH Hoenggerberg, CH-8093 Zuerich, Switzerland. E-mail: (akca, agruen)@geod.baug.ethz.ch 
KEY WORDS: Cultural Heritage, Image Analysis, Boundary-tracing, Fourier descriptors, shape similarity measures, Tree-Search, 
Visualization 
ABSTRACT: 
A stack of 66 historical Indian palm leaves, which were produced in the 8 th Century AD, is kept in the Museum Rietberg, Zuerich. 
On the leaves, there are figures and a long poem, inscribed in ancient Sanskrit language and narrating a love story. The original 
sequence of the leaves was lost long time ago. At one point in history, the stack of the leaves was damaged by a mouse biting pieces 
off. Only the first 18 leaves have their pages numbered in Sanskrit language, but the rest of them got out of order. If it is assumed 
that the mouse chewed at the leaves in a regular manner, the geometry of the leaf perimeter, as left over after eating, should bear 
useful information to find the original sequence. After acquiring digital images of the leaves and a pre-processing phase, an inner 
boundary-tracing algorithm was applied to all leaves in order to segment them. The fundamental data used in this work are boundary 
coordinates of the leaves. In order to obtain quantitative shape similarity measures, two different Boundary Based Shape Descriptor 
algorithms were applied to the boundary data: Fourier descriptors and a rotation-translation invariant boundary intersection-based 
shape descriptor. Shape descriptors indicate the similarity of different leaves. These similarity measures among all of the leaf pairs 
were arranged in form of a symmetric square matrix. With this matrix and a threshold similarity value one can determine the most 
probable ancestor and successor leaves for a pointed leaf. In the final step, a Tree Search scheme that starts from the 18 th (fixed) leaf 
and ends at the 66 th (relaxed) leaf was established to generate the most probable sequence. Every node in the tree was defined as a 
leaf and branched to the most probable neighbor leaves. The similarity measures were expressed as costs of the arcs, which connect 
two nodes in the tree. The sequence which has minimum total path cost was proposed as the most probable original sequence. 
1. INTRODUCTION 
In the collection of Museum Rietberg Zuerich is an old palm 
leave manuscript from India, consisting of 66 folios, inscribed 
on both sides. Originally, they had all the same dimensions and 
were once tightly bundled. On them are 100 erotic poems by a 
8 th century AD Sanskrit poet (by the name of Amaru) inscribed, 
many of which are illustrated with one or more pictures. The 
manuscript was prepared about 200 years ago to be kept as a 
bundle. All folios have a hole in the center through which a 
string was drawn to tie all leaves tightly between two wooden 
boards (Figure 1). However, the sequence of these poems is 
unknown (Fischer, 2002). The manuscript was kept for two 
centuries in an Indian library. Some time ago, a mouse did start 
to eat part of these folios. It is likely that at that time, the right 
sequence was still established, which was later lost, except at 
the beginning: Folios 1-18 are paginated and therefore their 
sequence is known. The mouse did not destroy the entire 
manuscript. She nibbled only about 5-10 % away from each 
leaf, mostly at the left side (Figure 2). If it is assumed that the 
mouse ate the leaves in a regular manner, the geometry of the 
leaf perimeter, as left over after eating, should bear useful 
information to find the original sequence. 
Since the rest of the leaves does not give any information about 
the sequence, we must focus only the geometry of the harmed 
part in order to avoid irrelevant data. The holes in the center of 
the leaves constitute a base point that is approximately at the 
same position on the all leaves. After the pre-processing phase 
i.e. rectification, the harmed left parts were cropped from the 
full images, and saved as different image files. All processes 
were performed based on these cropped images. 
After acquiring digital images of the leaves and a pre 
processing phase, an inner boundary-tracing algorithm was 
applied to all leaves in order to segment them. The fundamental 
data used in this work are boundary coordinates of the leaves. 
In order to obtain quantitative shape similarity measures, two 
different Boundary Based Shape Descriptor algorithms were 
applied to the boundary data: Fourier descriptors and a 
rotation-translation invariant boundary intersection-based 
shape descriptor. Shape descriptors indicate the similarity of 
different leaves. These similarity measures among all of the leaf 
pairs were arranged in form of a symmetric square matrix. With 
this matrix and a threshold similarity value one can determine 
the most probable ancestor and successor leaves for a pointed 
leaf. 
Figure 1: Front face of a leaf with hole for bundling up. Left perimeter side is damaged by the mouse.