Determination of Maximally Stable Extremal Regions in Large Images

J. Wassenberg, D. Bulatov, W. Middelmann, and P. Sanders (Germany)


Image Segmentation, Algorithm Engineering, Wide-Area Remote Sensing, Change Detection.


Coping with ever-increasing data requires efficient algo rithms. The topic of this work is segmentation; we present a new means of computing Maximally Stable Extremal Re gions that exhibits both high performance and low memory use and is thus able to process large scenes. An efficient data structure and improved variant of the Union-Find al gorithm make possible a software implementation that is competitive with an FPGA. After laying out the motivation and algorithm details, we derive a bound on the complexity and compare performance with that of other implementa tions. The topic is concluded with discussion of applica tions and future work.

Important Links:

Go Back