Latent Palmprint Image Segmentation Based on Dissimilarity Tolerance

The segmentation is the first and crucial step of the automatic forensic palm print system (AFPS) and other image process system. However, it is still manually cut in AFPS in the recent literatures. This paper proposes a segmentation algorithm with this principle: the segmentation is to separate the neighbor regions that run out the dissimilarity tolerance, which should be set according to the in trinsical character of the images. The latent palm print image segmentation function is composed of the intensity gradient and the frequency energy. Region growth is carried through a learning database of pure palm print blocks for the dissimilarity tolerance threshold. The mathematical morphology and other post-processes are performed to eliminate remnant fingers, the white center gap, and the noise. Experiment result substantiates the robust feasibility of the textual automatic segmentation algorithm in differently complex cases.