Abstract: High-resolution latent palmprint image is one of the primary forensic evidences. However the image is still manually segmented for its huge size and complex texture. Two automatic segmentation algorithms are proposed to solve this problem. The first detects the background based on the frequencies of ULBP in each block. The other compares the texture dissimilarity between neighbor regions in each couple until all the regions grow up into the foreground and background. The experiments on the standard latent palmprint database show effectiveness of the two algorithms.

Key words: palmprint segmentation, ULBP, region growing