Abstract
We present a fully automatic model based system for segmenting bone MR images of the knee. The segmentation method is based on a fast Active Appearance Models (AAM) based on canonical correlation analysis algorithm (CCA-AAM) where the dependency between texture residuals and model parameters are estimated in fast manner. The model is built from manually segmented examples from the knee images. The model has been applied to some challenging knee MR images. Experiments show that CCA-AAMs based segmentation, while requiring similar implementation effort, consistently outperform segmentation model based traditional AAM. Finally, we show results on knee image to illustrate the performance that are possible.