A scheme for detecting heterogenous regions in dermatological images with malignant melanoma is proposed. The scheme works without setting any parameter. The mean shift detection problem is divided into two stages: window size optimization and detection. In the first stage, the maximum circular neighborhood centered on each pixel for which it is true that all the elements belong to the same class as the central one is estimated using redundant data sets generated with overlapping groups. Statistics are computed from all these neighborhoods and associated ot the respective central pixels. As expected, larger values of a minimizing energy function are assigned to pixels belonging to heterogeneous regions. In the second stage, those regions are detected by applying first an expectation-maximization algorithm and, afterwards, automatically defining a threshold between homogeneous and heterogeneous regions. The present scheme is tested on a set of synthetical images. Results are shown on synthetical and real images. Extensions of the scheme to textural cases are also shown.
[1]
Y. Kareev.
Through a narrow window: working memory capacity and the detection of covariation
,
1995,
Cognition.
[2]
M. Moller,et al.
Supervised learning on large redundant training sets
,
1992,
Neural Networks for Signal Processing II Proceedings of the 1992 IEEE Workshop.
[3]
David A. Cohn,et al.
Active Learning with Statistical Models
,
1996,
NIPS.
[4]
Juan Lira.
Introducción al tratamiento digital de imágenes
,
1995
.
[5]
Hong Yan,et al.
Signal processing for magnetic resonance imaging and spectroscopy
,
2002
.