Text this: A Multi-View Semi-supervised learning method for knee joint cartilage segmentation combining multiple feature descriptors and image modalities