Level Set Segmentation Based on the Prior Shape of Biological Feature

The identification of user’s identity which is based on the analysis and measurement of the biological characteristics has become a research hotspot. The precise location and segmentation of the target is the basis for accurate biometric recognition. In view of the similarity of the external shape of human biological characteristics, the prior shape knowledge is introduced into active contour model based on level set. First, the training data of the shape function, which is expressed by the level set, are projected onto a lower dimensional subspace and achieved the primary attribute reduction on approximately Gaussian distribution of the training set using PCA method. Second, the further optimized properties are obtained by minimization class attribute interdependence minimization (CAIM) algorithm. Finally, under the constraints of the prior shape and object personality traits, level set curve based on the border and region can accurately evolve into the target boundary. Experiments demonstrate that our model can cope with image noise and clutter, as well as partial occlusions.