Translational calibration in region-of-interest localization for palmprint recognition

Palmprint is a promising biometric modality and has several advantages. However, many serious challenges severely restrict the development of palmprint recognition in mobile environments. Double-line-single-point (DLSP), as a state-of-the-art assisting graph, can be free from pre-processing, and effectively reduce the errors of pre-processing; however, its accuracy heavily depends on users’ cooperation and operation proficiency. This paper proposes boundary line calibration (BLC) and finger valley calibration (FVC) to suppress the translational dislocations in DLSP. Firstly, the samples acquired according to DLSP are rotated around the center so that the assisting lines are horizontal. Secondly, a rectangular area is cropped as the region of line detection (RLD) with the assisting line as the reference; while a rectangular area is cropped as the region of point detection (RPD) with the assisting point as the reference. The cropped rectangular regions are convolved with Gabor filters along an optimized direction to enhance edges. The accurate palm boundary is localized in RLD to reduce the vertical dislocation; and the accurate inter-finger valley point (IFVP) is localized in RPD to reduce the horizontal dislocation. The sufficient experiments show that the proposed algorithm can calibrate the translational dislocations caused by users’ imperfect cooperation, and accordingly remarkably improve the accuracy and users’ comfort.