Content-Based Retrieval of Aurora Images Based on the Hierarchical Representation - Advanced Concepts for Intelligent Vision Systems - page 258

A database of 129 images, which does not include the 40 sample images used to
derive the threshold values of the parameters for the classification purpose, was cre-
ated. There are 45 Type1 , 27 Type2 , and 37 Type3 aurora images. Some of the aurora
images belong to two types. These images are from the time period 1997 to 1999,
which are used in the experiment by Cao et al. [8] The database is searched for the
automatic retrieval of each type of aurora using the approach described previously.
Table 1 shows precision and recall of Type1 , Type2 , and Type3 aurora retrieval.

Table 1. Precision and recall of three types of aurora

Precision Recall

Nr Nt Nr / Nt Nr Nd Nr / Nd

Type1 42 42 100% 42 45 93.3%

Type2 23 26 88.5% 23 27 85.2%

Type3 37 37 100% 37 37 100%

A study of the misclassified images reveals that the misclassification was mainly
due to poor preprocessing. It is possible to achieve better accuracy by improving the
segmentation and preprocessing steps. In conclusion, the feasibility of building a
content-based image retrieval system based on the hierarchical representation is
demonstrated.

6 Conclusion

In this paper, we have presented the content-based retrieval system for aurora images.
The system utilizes the graph/tree hierarchical representation obtained from the boun-
dary based image segmentation and representation system and extracts various geo-
metric features for the purpose of classification. Those features include general shape
attributes of a curve, the convex hull, and the minimum bounding rectangle of an
object. The experimental results have proven that the hierarchical representation sup-
ports the fast and reliable computation of several geometric features and those geo-
metric features extracted directly from the hierarchical representation are useful for
content based image retrieval and also for a wide range of image interpretation appli-
cations. Other applications such as shape matching under rotation and scale changes
and recognition of the license plate can be found in [4].

References

1. Li, X., Ramachandran, R., Movva, S., Graves, S., Germany, G., Lyatsky, W., Tan, A.:

Dayglow Removal from FUV Auroral Images. In: Proceedings of IEEE International Geo-
science and Remote Sensing Symposium, vol. 6, pp. 3774-3777 (2004)

2. Li, X., Ramachandran, R., He, M., Movva, S., Rushing, J., Graves, S., Lyatsky, W., Tan,

A., Germany, G.: Comparing Different Thresholding Algorithms for Segmenting Auroras.
In: Proceedings of International Conference on Information Technology: Coding and
Computing, vol. 2, pp. 594-601 (2004)