Flexible Mixture Models for Colour Image Segmentation of Natural Images

This work proposes a way to increase the flexibility of mixture models for clustering applications. The focus of this paper resides at the segmentation of natural images, as one application of mixture models in real world applications. Furthermore, a novel method for estimating the number of mixture components utilising a Poisson regression approach is proposed. The derived procedure is applied and evaluated on the Berkeley Segmentation database (BSD500) [1] and on the LeafSnap Field dataset [2], a challenging real world application of segmentation tasks in the domain of plant recognition.