CI PA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
59 
In the implementation phase, the number of allowed successor 
leaves for an ancestor leaf was defined according to two 
predefined rules (limits): 
• Limit T c : maximum count for allowed nodes 
• Limit T v : maximum value limit for shape similarity 
measure from matrix Df ouner or ¿^intersection 
These two limits state that the number of successor nodes for an 
ancestor node cannot exceed the limit T c , and the cost value 
for any arc cannot exceed the limit T v . In tree search methods 
search time increases exponentially with the number of used 
combinations. These two limits reduce the combinations. Some 
leaves in the set have completely different shapes with respect 
to other leaves. This means that either none of the leaf might 
give a branch (arc) to these different shaped leaves or these 
extremely different leaves might not find similar leaf (leaves) as 
successor. To get rid of this problem one should relax the limits 
T c and T y . In this case, the search time increases 
exponentially. Therefore the leaves are splitted into groups to 
solve the problem in an optimum way. According to their 
shapes, the leaves were grouped into four categories. Each 
group was ranked internally and then all the groups were 
merged. The obtained result is given in the following Figures 
(11), (12), and (13). 
3. CONCLUSIONS 
In this work, the missing sequence of the historical palm leaf 
manuscript was found by using shape descriptors. Two 
boundary-based shape descriptor algorithms were applied. Both 
of these shape description algorithms can find the most similar 
neighbour leaf for a pointed leaf successfully. The fundamental 
problem of this work is to establish the original sequence for all 
leaves using shape similarity information provided by the shape 
description algorithms. A tree search scheme was performed to 
achieve this task. The sequence, which has the minimum total 
cost, was proposed as the most probable sequence. An 
alternative method to the tree search method might be 
probabilistic relaxation. In this case, similarity measures 
provided by the shape description algorithms can be interpreted 
as probability of neighborhood. In order to provide a useful 
visual inspection for the proposed sequence, a texture mapped 
model of palm leaf manuscript was visualized and animated. 
ACKNOWLEDGEMENT 
This project was supported by Museum Rietberg, Zuerich. The 
cooperation and help of Dr. Eberhard Fischer is gratefully 
acknowledged.