Non-rigid Object Segmentation Using Robust Active Shape Models

Statistical shape models have been extensively used in several image analysis problems, providing accurate estimates of object boundaries. However, their performance degrades if the object of interest is surrounded by a cluttered background, and the features extracted from the image contain outliers. Under these assumptions, most deformable models fail since they are attracted towards the outliers, leading to poor shape estimates. This paper proposes a robust Active Shape Model, based on a sensor model that takes into account both valid and invalid observations. A weight (confidence degree) is assigned to each observation. All the observations contribute to the estimation of the object boundary but with different weights. The estimation process is recursively performed by the Expectation-Maximization method and the weights are updated in each iteration. The algorithm was tested in ultrasound images of the left ventricle and compared with the output of classic Active Shape Models. The proposed algorithm performs significantly better.