Abstract:

We proposed an improved image segmentation model based on Local Binary Fitting (LBF) to solve the difficulties in finger vein image segmentation caused by fuzzy localness, low contrast and inhomogeneous intensity. The model inherits the advantages of LBF model in selecting Gaussian kernel to acquire local information, which is very effective for segmenting image of inhomogeneous intensity. Specially, a weighted total variant component with edge stopping function is added to the improved model, which gives the model higher sensitiveness to edge information and better performance on segmenting image with low contrast. Level set regularization is introduced to the model to avoid re-initialization of the level set function, thus reducing the evolution time. Experiments on simulated data and real finger vein images indicate that the improved model is effective for finger vein image segmentation, which has higher segmentation capacity with regard to detailed edge information. Also it is less possible for the improved model to fall in local convergence.

Key words: information processing, image segmentation, active contour model, finger vein image, local binary fitting(LBF), weighted total variation