Semi-supervised learning spectral embedding with active contours model

We present a semi-supervised learning algorithm to recognize feature vector noises in the training data. Our proposal employs an active contour model technology (ACM) which is used for objects extraction in the field of computer vision. We extend the ACM technology to the similarity formula of our proposal for identifying feature vector noises in the training set and improve the performance of the training data. The proposal is applied to the synthetic data and real data. The experiments prove that the proposal has a high performance on the feature vector noises in the unlabeled data of the training set.