Object Segmentation
Kluwer Academic Publishers eBooks · 2006
We have highlighted some experiment results on home photo segmentation. The MRF/GD model is proved to be useful in practice. Experiment results demonstrate that the CIE— L*u*v* color space performs better than HSV in the unsupervised clustering based segmentation on digital home photos. The region merging algorithm can serve as an effective compensation step for the clustering based segmentation. Texture features extracted by the wavelet frame transform are found to be promising in discriminating different textures. The combination of color and texture for the segmentation of natural images is a challenging research area.