bul 2004 
ithm here 
ary pixels 
t types of 
| way, the 
4. Conclusions 
In this paper, we present a novel approach for multi-scale 
segmentation of images, which is based on granulometry and 
anti-granulometry. Through derivatives within an image 
sequence, we can get the morphological feature of the image on 
which the image segmentation is performed. In this method we 
presume that pixels with same morphology features belong to 
the same object. Experiments show that the proposed approach 
is more effective than the watershed method. The problems of 
the boundary pixels and over segmentation are avoided because 
the algorithm is based on the regional feature. This method fits 
to segments of remotely sensed images with higher spatial 
resolution. 
From this study, the following problems still need to be further 
addressed. 
Selection of structure element: The objects in reality are 
complicated and have various shapes so that the structure 
element can be selected in many ways. The selection of suitable 
structure elements is a key step for analyzing morphological 
features. It is impossible for structure elements with a single 
shape and variable sizes can suit all objects in the image. We 
should study whether a multi-shape structure element sequence 
can be used. 
The morphological feature of every pixel was assumed to be 
unique in the algorithm of this paper. When the image possesses 
complex objects, some pixels may have more than one 
significant derivative maximum, i.e., morphological feature of 
every pixel is not unique. In this case, it is a problem to classify 
the morphological feature of the pixel. 
Selection of parameter © under soften condition: The parameter 
O reflects the sensitivity of objects to structure elements. If 
the parameter is suitable, we can obtain good results of image 
segmentation under a certain sequence of structure elements. 
We need to further study the relations among shape and size of - 
structure elements, parameter © and image. 
The morphological feature obtained by the algorithm proposed 
in this study reflects the structure and size of objects in the 
image. The shape and size of objects in the resultant image 
would be influenced heavily by the structure elements. So the 
resultant image may provide some false information if the 
51 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004 
structure elements are not selected properly. A possible 
improvement is to consider segmenting images through cluster 
vectors formed by morphological and other image features. 
Aknowledgements 
This work was funded by a 937 program (Grant No. 
2003CB415205) of the National Natural Science Foundation of 
China (Grant No. 40176032), Hong Kong Polytechnic 
University (Grant No. A-PF54 ) and the Opening Foundation of 
LED, South China Sea Institute of Oceanography, Chinese 
Academy of Sciences. 
References 
Beucher. S., 1994, Watershed, Hierarchical Segmentation and 
Waterfall Algorithm In: Proc. Mathematical Morphology and 
Its Applications to Signal Processing, Paris, France, pp. 69-76 
Chanussot. J. and P. Lambert, 1998, An Application of 
Matheatical Morphology to Road Network Extraction on SAR 
Image. In: Proc. Mathematical Morphology and its Applications 
to Signal Processing ,Amsterdam,The Netherlands, pp. 
399-406 
Cui, Y., 1999, Image Process and Analyse — Method and 
Application of Mathmatical Morphology Science Press, Beijing, 
pp. 1-14 
Demarty, C. and S. Beucher, 1998, Color Segmentation 
Algorithm Using an HLS Transformation. In: Proc. 
Mathematical Morphology and its Applications to Signal 
Processing, Amsterdam, The Netherlands, pp. 2312-238. 
Geraud. T, P-Y.Strub, and J. Darbon, 2001, Color Image 
Segmentation Based On Automatic Morphological Clustering, 
In: Proceedings of the IEEE International Conference On 
Image Processing. Vol.3. Thessaloniki, Greece, pp. 70-73 
Katartzis, A. V. Pizurica and H. Sahli, 2000, Application of 
Mathematiocal Morphology and Markov Random Field Theory 
to the Automatic Extraction if Linear Features in Airborne 
Images. In: Proc. Mathematical Morphology and its 
Applications to Signal Processing. California, USA, pp. 
405-414