A new method for building adaptive Bayesian trees and its application in color image segmentation

This paper presents a novel non-supervised clustering method based on adaptive Bayesian trees (ABT). A Bayesian framework is proposed for seeking modes of the underlying discrete distribution of the input data, and the data is represented by hierarchical clusters found using the adaptive Bayesian trees approach. The application of the proposed clustering technique to color image segmentation is investigated, exploring the inherent hierarchical tree structure of the proposed approach to represent color images hierarchically. The experimental results with the BSD300 dataset and 21 comparative methods that are representative of the art suggest that the proposed ABT clustering scheme potentially can be more reliable for segmenting color images than the comparative approaches. The proposed ABT approach achieved an average PRI value of 0.8148 and an average GCE value of 0.1701, suggesting that potentially the proposed scheme can improve over the comparative methods results. Also, the visual evaluation of the results confirm the competitiveness of the proposed approach. Other applications of the ABT clustering scheme in computer vision and pattern recognition currently are under investigation.