Smart touchless palm sensing via palm adjustment and dynamic registration

Abstract

Touchless palm recognition is increasingly popular for its effectiveness, privacy, and hygiene benefits in biometric systems. However, several challenges remain, including significant performance degradation caused by variations in palm positioning and capture distance. To address these issues, this paper introduces a comprehensive sensing system that integrates dynamic registration with robust palm adjustment. Specifically, we conduct a thorough investigation of distance variations to establish optimal registration settings. In addition, we propose an edge-aware, rotation-invariant region of interest alignment method, which ensures spatial alignment for any given palm across its different samples, even under challenging conditions. By embedding it into a palm registration framework based on video sequences, we improve the system's ability to adapt to varying conditions automatically. Extensive experiments on various datasets demonstrate that the proposed method significantly enhances the performance of touchless palm recognition systems.

Fig. 1. Contactless palm sensing and corresponding challenges.

a Image sharpness and brightness variations caused by different placement distances. b Palm pose variations and ROI alignment. Here, p_t1 and p_t2 are the keypoints of the two finger valleys, respectively; point c is the center of the palm. c The distance and lighting factors present in palm sensing systems. d The multi-range palm registration process.

Fig. 2. Trend of recognition performance as the number of registered samples per palm increases, for different coefficients.

The default settings for the coefficients are: k_a = 0.12, k_d = 0.15, k_s = 0.10, and t_q = 0.03, where k_a, k_d, and k_s are the coefficients for ambient, diffuse, and specular reflections, respectively, and t_q is the light attenuation parameter. EER is an abbreviation for Equal Error Rate. a–c show the trend for varying ambient coefficient k_a values of 0.10, 0.14, and 0.18, respectively. d–f display the trend for different diffuse coefficient k_d values of 0.10, 0.20, and 0.30, respectively. g–i illustrate the trend for varying specular reflection coefficient k_s values of 0.05, 0.15, and 0.25, respectively. Finally, (j)–(l) present the trend for different quadratic parameter t_q values of 0.02, 0.04, and 0.06, respectively.

Fig. 3. Alignment results for palm samples with varying positions from a single palm.

The first row shows the original palm images. The second row displays the ground truth alignment. The remaining rows present the alignment results from Ours, ColScan, RefPoint, ConHull, BaoNet, IzadNet, and PKLNet, respectively. In the figure, the red rectangles are the ROI localization results of each method, and a cross mark ( × ) in the top-left corner of an image indicates the failure of a method to detect any points.

Fig. 4. Imaging model for palmprint sharpness and brightness analysis at different palm distances.

a Point p is imaged in front of the CMOS plane. b Point p is imaged on the CMOS plane. c Point p is imaged behind the CMOS plane. d LEDs arrangement. (e) Light intensity at point p. f Lighting model. g A typical palm image acquisition device. Here, l_d and l_f are the diameter and focal length of the camera lens. V_n represents the normal vector of the reflection point’s local plane, while V_l, V_r, and V_v are unit vectors corresponding to the directions of the incident light, reflected light, and viewer’s perspective, respectively.

Fig. 5. Robust contactless palmprint ROI extraction.

a Pipeline of the ERAlign method. After rough rectification, a palm may still has small angular deviations that require subsequent steps for fine-grained alignment. b Coordinate system for hand ROI localization. Here, x-o-y and $x^{\prime }$-$o^{\prime }$-$y^{\prime }$ are the palm and image coordinate systems, θ is the angle between the two coordinate systems. p_t1 and p_t2 are the finger valley points, and p_r1-p_r4 are the four corner points of the ROI. l_r is the length of the tangent. t₁ and t₂ are parameters to respectively adjust the ROI's translation and scale.