Spinal Images Segmentation Based on Improved Active Appearance Models

Active Appearance Models (AAMs) is a deformable model based on statistical information, and also is an efficient method of image segmentation by extracting the features of the image. Its statistical analysis is Principal Component Analysis (PCA). PCA only takes into account the second order statistical information, which doesn´t include the phase information, so it is difficult to extract the local features. In order to overcome this problem, Independent Component Analysis (ICA) is proposed to improve the original AAMs in this paper. The eigenvectors that yielded by PCA describe global variations, while the vectors yielded by ICA describe local variations, thus ICA shows a stronger ability to describe local features than PCA. In this paper, a new approach of spinal images segmentation based on improved AAMs is presented, and the experimental results demonstrate that our method outperforms standard AAMs.