Robust ROI localization based on image segmentation and outlier detection in finger vein recognition

Finger vein is deemed to be a promising biological trait for individual identification. However, partially due to non-uniform collection devices and non-standard collection process, original images are polluted by lots of unfavourable factors. These negative effects increase the burden on image matching. Therefore, Region of Interest (ROI) localization plays an important role in finger vein recognition. Considering that the previous techniques are not common for all kinds of images, we propose a set of methods to obtain the ROI, which is able to remove most of negative factors, preserve more vein information and keep the stability of vein feature with less cost and fewer manual thresholds. More specifically, we propose Simplified Statistical Region Merging (SSRM) with dynamical adjustment of precision parameter to segment an image into finger body and background area. Next, in order to ensure the edge be qualified and further correct the skew angle, the novel Directional Linkage Clustering Method (DLCM) and Parameter Selection (PS) are introduced. Compared with the previous work, the number of thresholds used during the whole process is reduced to only four. The identification EER in experiments is reduced to 0.0476 on all the images in three public databases, which indicates that our method is more superior than the compared methods and performs better in the individual identification.