IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring”, Hyderabad, India,2002 | 
are productive and important for resources exploration. In the 
surface profile it gives appearance of mender with curvature 
covering large aerial extend and in spatial profile there is gradual 
increase in concentration as the distance increases. 
Sharp linear front: Fronts with high magnitude of chlorophyll 
gradient. It appears sharp boundary between the two different 
types of water masses. There is abrupt increase in concentration 
and sudden decline in concentration is seen in surface profile. 
There is sharp rise in concentration for short distance and sudden 
decline in concentration is clearly observed in spatial profile. 
Clear water: The water mass with low concentration of 
chlorophyll. The clear water or areas without features in the 
offshore regions indicates almost no change in the concentration 
as seen in surface and spatial profiles. There is no variation in the 
profiles. 
3.2. Feature extraction 
Figure 3 exhibits the subset of chlorophyll image showing 
different types of features and out puts of application of different 
mathematical algorithms for texture analysis on this subset. The 
features types like rings, eddies, meanders 
  
Figure 3: Feature extraction using texture analysis algorithm. (a). 
Subset of chlorophyll image (b). Skew image (c). Mean image (d) 
Variance image. 
and fronts are clearly extracted in variance image as compared to 
mean and skew image. This indicates that variance algorithm is 
most suitable than mean algorithm. It appears that skew algorithm 
is less suitable as many features are not detected in skew image. 
3.3. Concept of pattern recognition 
Pattern recognition usually denotes classification and / or 
description of a set of processes or events. Sets of events or 
processes with some similar properties are grouped into the 
classes. The total number of patterns in a particular problem is 
often determined by the particular application in mind. Pattern 
recognition is one of the central problems in the design of many 
automated decision making systems. A pattern recognition system 
typically acquires input data using sensors, a representation of 
acquired data is obtained using feature extraction algorithm and 
finally a decision is made based on the feature vector. The 
decision-making component of a pattern recognition system is 
commonly manifested in two distinct forms, they are matching 
and classification. An intuitively appealing approach for pattern 
recognition is "template matching". A set of templates or 
prototypes, one for each pattern class is stored in the 
device/machine. The input pattern (with unknown classification) 
is compared with template of each class, and classification is 
based on a pre-selected matching criterion or similarity criterion. 
The matcher rejects/accepts the hypothesis of whether the patterns 
are same or not. A classifier performs the categorization. 
The patterns of surface profile in three-dimension space and 
spatial profile in two-dimension clearly give overall idea of the 
type of features present in an image, changes in the features, the 
gradient of concentration and the morphology. The vector of the 
features can be generated to create the templates of each type of 
feature for pattern recognition. The templates of each feature may 
be stored in the machine and can be utilised for template 
matching. Artificial neural net (ANN) work analysis is suggested 
for the feature selecting using pattern recognition system. The 
following steps are suggested for developing objective technique 
to locate PFZs automatically. 
1. Sets of templates for matching the input pattern may be 
generated and stored in the machine. 
2. Feature extraction and vector generation. 
3. Classification of input pattern through "matching 
templates" or other method. 
4. Multi-date Monitoring of pattern to 
persistence of oceanic features. 
5. Classification and categorization of feature as per the 
gradient, shape, quantity i.e. chlorophyll concentration 
and persistency. 
understand 
4. CONCLUSIONS 
Surface and spatial profiles of oceanographic features exhibited 
their shape, concentration and morphological patterns. A variance 
algorithm was found most suitable for feature extraction. These 
clues can be utilized for developing objective technique for 
locating PFZs. The templates of each feature types may be 
generated for template matching in the pattern recognition system. 
Artificial neural net (ANN) work analysis is suggested for the 
feature selecting using pattern recognition system. 
Acknowledgements 
The authors express their sincere gratitude to Dr. A.K.S. Gopalan, 
Director, Space Applications Centre. We are also thankful to Dr. 
M.P. Oza and Mr. K.K. Mohanty for useful discussion. 
467