-
指纹识别技术在安保、门禁等领域已较为成熟,但与专用指纹采集设备不同,案发现场指纹图像一般质量较低、对比度较低,直接识别可能出现错误,需要进行一定的前处理以增强图像。由于现场指纹情况复杂,难以针对性设计,本文中根据指纹特点提出一种自适应增强算法,包括多光谱图像融合增强及指纹图像增强两个步骤。
-
多光谱图像增强技术已广泛应用于遥感探测领域。由于本装置可获得不同波段光源照明时的指纹图片,因此, 借鉴该技术能够充分利用指纹在不同波段处反射率差异增加图像对比度。目前已有多种多光谱图像增强算法,本文中采用主成分分析(principal component analysis,PCA)技术。具体流程如下。
(1) 由于智能手机一般拍摄为彩色图像,需将其变换为灰度图像,变换公式如下式所示:
$ g=0.2989R+0.5870G+0.1140B $
(1) 式中, R, G及B分别为原图像红色、绿色及蓝色分量,g为变换后灰度值,假设共拍摄N波段多光谱图像,强度记为Ii(i = 1, …, N),行数及列数分别记为r及c。
(2) 将每个波段2维图像矩阵转成1维向量,获得图像向量矩阵M,其行数与列数分别为r×c及N,求解M的协方差矩阵C及其特征向量矩阵V与特征值对角矩阵A,提取最大特征值Am及对应特征向量Vm。
(3) 融合增强后图像如下式所示:
$ \boldsymbol{I}=\boldsymbol{M} \cdot \boldsymbol{V}_{\mathrm{m}} $
(2) -
研究人员已开发了多种指纹图像增强算法,如灰度均衡、小波变换、卷积神经网络等,这些算法各有优缺点及适应场合,本文中选择改进依据指纹本身特点进行增强的方向滤波算法,使用与指纹局部纹线方向一致的滤波器,增强实用性,具体流程如下。
(1) 方向图计算。指纹图像在纹线方向变化缓慢,垂直纹线方向则变化剧烈,因此纹线方向与图像梯度密切相关。本文中使用Sobel算子计算图像梯度,如下式所示:
$ \left\{\begin{array}{l} \boldsymbol{G}_{\mathrm{i}}=\sqrt{\boldsymbol{G}_{x}^{2}+\boldsymbol{G}_{y}^{2}} \\ \boldsymbol{G}_{x}=\left|\begin{array}{ccc} -1 & 0 & 1 \\ -2 & 0 & 2 \\ -1 & 0 & 1 \end{array}\right| * \boldsymbol{I}_{0} \\ \boldsymbol{G}_{y}=\left|\begin{array}{ccc} -1 & -2 & -1 \\ 0 & 0 & 0 \\ 1 & 2 & 1 \end{array}\right| * \boldsymbol{I}_{0} \end{array}\right. $
(3) 式中,*表示卷积,I0是需要求解梯度的原图像。由于梯度方向相反时方向图应保持一致,因此将方向角定义在第一象限和第四象限,如下式所示:
$ \theta=\arctan \left(\boldsymbol{G}_{y} / \boldsymbol{G}_{x}\right) $
(4) (4) 式表明方向角范围为(-π/2, π/2],进一步对其进行离散化处理,将方向角范围均匀离散至16个区间,即(-π/2, -7π/16], (-7π/16, -6π/16], …, (7π/16, π/2],并新建数组保存方向角直方图。由于指纹图像在局部方向图变化较为缓慢,因此为减小计算量,将图像进行分块,尺寸为16×16,遍历图块各像素,建立直方图,选择具有最多像素的方向角作为图块整体方向,最终获得方向图。
(2) 方向滤波器设计。方向滤波器一方面增强沿着指纹纹线方向的点并减弱随机噪声影响,另一方面断开垂直纹线方向粘连以提高指纹质量。假设图块中心像素坐标为(0, 0),对应方向角为θ0,则过图块中心的方向直线可表示为:
$ x \sin \theta_{0}+y \cos \theta_{0}=0 $
(5) 根据定义,方向滤波器应在平行直线的方向具有相近权重,垂直直线的方向具有渐变权重,因此采用距离函数定义滤波器元素权值为:
$ W=A_{0}-A_{1}^{D} $
(6) 式中, A0及A1为经验参量,分别设置为11及3;D是指数,表示该元素位置与过图块中心的直线之间的距离,单位为像素。
$ D=\left|x_{0} \sin \theta_{0}+y_{0} \cos \theta_{0}\right| $
(7) 式中, (x0, y0)为元素的像素位置。由于方向角已离散为16个数值,方向滤波器可预先计算并存储于内存中,后续滤波时直接根据方向角调用相应滤波器与图块进行卷积运算即可。
完成滤波器设计后,分别对每块图像进行卷积滤波即可实现图像增强。基于上述原理,本文中的算法流程如图 3所示。
-
对搭建的指纹图像自动化采集装置进行了实验研究。首先在蓝色背景颜色纸张上使用红色印泥制作了指纹,进一步测试了多光谱采集及图像增强的效果。为了模拟现场实际指纹质量可能较差的情况,实验中并未特别注重指纹质量,指纹存在较严重的粘连。由于实验拍摄时只需固定好装置,LED为电子控制,无需任何机械调节,因此图像采集具有较高的效率,4幅图像约需0.5s。
图 4为4种LED照明时拍摄的典型指纹图像。由图 4a~图 4d分别为紫色、蓝色、红色、绿色LED照明时采集图像的灰度图。红色LED照明时蓝色背景强度较弱,部分红色粘连指纹反射较强,显示为白色,但左侧及下方指纹反射很弱,对比度较差,表明红色照明难以适用于该指纹的图像采集需求。蓝色与紫色LED的照明波长较为接近,采集的图像相近。与预期一致,指纹主体部分反射率低于背景部分,因此在图像上显示为黑色指纹,但值得注意的是,指纹中部及下方部分纹线之间背景为白色,这主要是因为制作指纹的纹线具有3维曲面面形,在特定角度光源照明时将光线镜面反射至纹线中间的背景上,导致背景反射强度较高,表明光源入射角度对成像效果有较大的影响。由于智能手机镜头一般具有一定的变焦能力,因此在实际测量时,可移动装置,通过控制其与指纹的距离调节光源入射角度。相比之下,图 4d中绿色LED照明时蓝色背景较亮,红色指纹纹线较暗,具有更高的对比度,因此绿色照明更为适合实验中指纹及背景颜色。
对图 4中多光谱图像进行了融合及增强测试,服务器使用普通个人计算机进行图像处理,CPU为Intel(R)Core(TM)i7-7700@3.60GHz,内存为8.0GB,图像像素数目为384×288,此配置下主成分分析法融合约需100ms。融合结果如图 5a所示。由于图 4d具有最高对比度,其对融合图像主成分影响最大,因此融合图像整体分布与图 4d相近,但图 4b中部及下方明亮的背景部分在融合图像中亦有所体现,表明主成分分析法能够将每幅图像的局部特征进行融合。进一步进行了方向滤波测试,对其进行分块处理,每块的像素为32×32,滤波器尺寸为7×7,滤波时间约需100ms,可满足一般场景处理效率需求,若需要融合及增强海量指纹图像,则可进一步通过并行计算提高效率。图 5b为进行方向滤波增强后的指纹图像。与原图像相比,图像具有更高的对比度,可为后续识别提供更高质量的指纹图像。同时, 尽管图像进行了分块,但并未观察到马赛克效应,表明设计的图像增强算法达到了预期的效果。
为了进一步定量研究算法性能,引入加权指纹图像质量评价指标,包括平均灰度指标、灰度模糊指标、全局方向一致性及全局脊线纹理清晰度。具体参量为:灰度阈值为[0.2, 0.8],灰度方差阈值为0.2,方向一致性阈值为0.2,纹线清晰度阈值为[5,10,4]。4种指标权值相同,指标详细解释及参量意义见参考文献[21]。计算结果如表 1所示,可见照明波长不同时,指纹图像质量变化明显。经过多光谱融合处理后,指纹图像质量相比最优波长有所改善,而在方向滤波之后大幅增加,表明多光谱融合及方向滤波能够有效改善指纹图像质量,为后续指纹识别提供了良好的基础。表中图像质量为纯数值计算结果,灰度范围值为0~1。
基于智能手机的多光谱指纹图像采集研究
Acquisition of multi-spectrum fingerprint image with a smartphone
-
摘要: 为了解决现场指纹图像采集过程中指纹背景颜色复杂、导致对比度和采集效率较低的问题, 采用智能手机设计了便携式指纹图像采集装置。一方面将手机连接单片机, 电动切换不同波长光源, 获得多光谱图像; 另一方面自动采集不同光源下的指纹图像。为确保指纹图像质量满足后续识别需求, 采用多光谱融合及方向滤波技术对图像进行了增强处理。结果表明, 当光源数目为4时, 采集及处理时间低于1s, 满足现场采集需求; 装置能够有效采集指纹图像并融合增强。该装置具有低成本、自动化及便携性高的优点, 有望在刑侦领域取得广泛应用。Abstract: In order to overcome the problem that the image contrast and collection efficiency of on-site fingerprint images are usually quite low due to complicated background color at the crime scene, a portable fingerprint image acquisition system with a smartphone was proposed. In the system, a microprocessor connected to a smartphone through bluetooth can electrically switch light sources of different wavelengths. Multi-spectrum images were acquired to improve the fingerprint image contrast. With the help of the microprocessor, the smartphone can automatically capture fingerprint images under the illumination of different light sources. To ensure that the image quality can meet the requirements of identification process, the images were further enhanced by using multi-spectrum fusion and orientation filtering. Results show that the proposed system can effectively collect, fuse, and enhance multi-spectrum fingerprint images. It takes less than 1s to acquire and process the images when the number of light sources is 4, which satisfies the requirement of on-site image acquisition. As the system has the advantages of low cost, automation, and high portability, it is expected to be widely applied in criminal investigation.
-
Key words:
- image processing /
- fingerprint image acquisition /
- multi-spectrum image /
- smartphone
-
-
[1] HAZARIKA P, RUSSELL D A. Advances in fingerprint analysis[J]. Angewandte Chemie International Edition, 2012, 51(15): 3524-3531. doi: 10.1002/anie.201104313 [2] TIEDGE T M, MCATEE P D, McCORMICK M N, et al. Massively parallel sequencing and STR analysis from partial bloody fingerprints enhanced with columnar thin films[J]. Forensic Science International: Genetics, 2020, 49: 102369. doi: 10.1016/j.fsigen.2020.102369 [3] SINGLA N, KAUR M, SOFAT S. Automated latent fingerprint identification system: A review[J]. Forensic Science International, 2020, 309: 110187. doi: 10.1016/j.forsciint.2020.110187 [4] WOLSTENHOLME R, FRANCESE S, BRADSHAW R. Study of li-pid distribution and degradation in latent fingerprints by spectroscopic imaging techniques[J]. Science & Justice, 2010, 50(1): 37-38. [5] LEINTZ R, BOND J W. Can the RUVIS reflected UV imaging system visualize fingerprint corrosion on brass cartridge casings postfiring[J]. Journal of Forensic Sciences, 2013, 58(3): 772-775. doi: 10.1111/1556-4029.12101 [6] CRANE N J, BARTICK E G, PERLMAN R S, et al. Infrared spectroscopic imaging for noninvasive detection of latent fingerprints[J]. Journal of Forensic Sciences, 2007, 52(1): 48-53. doi: 10.1111/j.1556-4029.2006.00330.x [7] TAHTOUH M, DESPLAND P, SHIMMON R, et al. The application of infrared chemical imaging to the detection and enhancement of latent fingerprints: Method optimization and further findings[J]. Journal of Forensic Sciences, 2007, 52(5): 1089-1096. doi: 10.1111/j.1556-4029.2007.00517.x [8] GHARBIA R, HASSANIEN A, EL-BAZ A, et al. Multi-spectral and panchromatic image fusion approach using stationary wavelet transform and swarm flower pollination optimization for remote sensing applications[J]. Future Generation Computer Systems, 2018, 88: 501-511. doi: 10.1016/j.future.2018.06.022 [9] ZHANG M, LI S, YU F, et al. Image fusion employing adaptive spectral-spatial gradient sparse regularization in UAV remote sensing[J]. Signal Processing, 2020, 170: 107434. doi: 10.1016/j.sigpro.2019.107434 [10] SUN G, ZHANG X, JIA X, et al. Deep fusion of localized spectral features and multi-scale spatial features for effective classification of hyperspectral images[J]. International Journal of Applied Earth Observation and Geoinformation, 2020, 91: 102157. doi: 10.1016/j.jag.2020.102157 [11] WENG S, TANG P, YUAN H, et al. Hyperspectral imaging for accurate determination of rice variety using a deep learning network with multi-feature fusion[J]. Spectrochimica Acta, 2020, A234: 118237. [12] SHARMA R P, DEY S. Two-stage quality adaptive fingerprint image enhancement using fuzzy C-means clustering based fingerprint quality analysis[J]. Image and Vision Computing, 2019, 83/84: 1-16. doi: 10.1016/j.imavis.2019.02.006 [13] GUPTA R, KHARI M, GUPTA D, et al. Fingerprint image enhancement and reconstruction using the orientation and phase reconstruction[J]. Information Sciences, 2020, 530: 201-218. doi: 10.1016/j.ins.2020.01.031 [14] KHAN T M, KHAN M A U, KONG Y. Fingerprint image enhancement using multi-scale DDFB based diffusion filters and modified Hong filters[J]. Optik, 2014, 125(16): 4206-4214. doi: 10.1016/j.ijleo.2014.04.048 [15] ULEP T, ZENHAUSERN R, GONZALES A, et al. Smartphone based on-chip fluorescence imaging and capillary flow velocity mea-surement for detecting ROR1+ cancer cells from buffy coat blood samples on dual-layer paper microfluidic chip[J]. Biosensors and Bioelectronics, 2020, 153: 112042. doi: 10.1016/j.bios.2020.112042 [16] SHARMA J, ONO T, SANDHU A. Smartphone enabled medical diagnostics by optically tracking electromagnetically induced harmonic oscillations of magnetic particles suspended in analytes[J]. Sensing and Bio-Sensing Research, 2020, 29: 100347. doi: 10.1016/j.sbsr.2020.100347 [17] WEI X, LI Ch L, XUE W, et al. Smart phone recognition characteristics of laser marking barcodes on aluminum alloy surface[J]. Laser Technology, 2016, 40(5): 633-637(in Chinese). [18] CALEB J, ALSHANA U, ERTAS N. Smartphone digital image co-lorimetry combined with solidification of floating organic drop-dispersive liquid-liquid microextraction for the determination of iodate in table salt[J]. Food Chemistry, 2021, 336: 127708. doi: 10.1016/j.foodchem.2020.127708 [19] PENG B, ZHOU J, XU J, et al. A smartphone-based colorimetry after dispersive liquid-liquid microextraction for rapid quantification of calcium in water and food samples[J]. Microchemical Journal, 2019, 149: 104072. doi: 10.1016/j.microc.2019.104072 [20] LI B, YU Y, CAO Y, et al. Point-of-care testing for streptomycin based on aptamer recognizing and digital image colorimetry by smartphone[J]. Biosensors and Bioelectronics, 2018, 100: 482-489. doi: 10.1016/j.bios.2017.09.028 [21] HAN Zh, LIU C P. Automatic evaluation algorithm of fingerprint images quality[J]. Computer Engineering, 2007, 33(19): 4-6(in Chinese).