RFPIQM: Ridge-Based Forensic Palmprint Image Quality Measurement

Forensic palmprint recognition is a biometric technique that identifies or verifies person identity with their palmprint images of resolution more than 500 dpi. Since the palmprint images with poor quality are inevitable and have a significant effect on every stage of the identification, it is valuable to measure the quality of the palmprint images. The previous work on high-resolution palmprint image quality measurement has some limitations. It lacks attention of latent palmprint and does not take full advantage of the properties of palmprint ridges. So only private, uni-modal, and small data set is used, and the classification accuracy needs to be improved. In this paper, we propose a general method to measure the image quality of a block or a full image from forensic palmprints based on the ridge properties. We first propose two new features (i.e., ridge period and ridge orientation variance) to measure the palmprint image quality. We also bring in some previous features, ridge orientation continuity, ridge thickness uniformity, and ridge-valley contrast, to enhance the classification performance. Then, we propose a supervised learning method to measure the quality of the palmprint images. We use labeled palmprint images training three different kinds of existing classifiers and then use them to predict the quality of the images. To show the reliability and stability of our method, cross validation is used on multi classifiers. And the comparison shows that our method has a high accuracy with fewer running time than the previous method. Furthermore, the experiment also shows that when our palmprint image quality measurement method is used to filter the unreliable minutiae, the matching accuracy is evidently improved.