Incidencia de la segmentación en la obtención de región de interés en imágenes de palma de la mano

Abstract

One of the greatest difficulties in pattern recognition and in particular biometric systems based on palmprint, is to obtain at preprocessing stage the adequate calculation of the target region (Region Of Interest, ROI), due to it directly influences the final results of the identification system. This paper presents a
comparison among images that contain the region of interest of palmprints. ROIs are extracted using two different segmentation techniques for each database. Otsu and Hang methods are used for the PolyU database, while Otsu and ISODATA methods are used for the CASIA database. To determine the similarity between the ROI extracted from each segmentation method, a comparison test is conducted using the following similarity measures: correlation, mutual information, mean square error, ratio image uniformity
and common area between images. This research showed that indeed the change of the segmentation methods generate changes in the ROI of the palmprint, as the location of the valley points change depending on the method. Finally, it concludes that when vary the threshold method, the images generated by each method are different and vary the method of obtaining of the ROI variation is greater.