Fig 1 - available via license: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Content may be subject to copyright.
Source publication
Iris Recognition (IR) is one of the market's most reliable and accurate biometric systems. Today, it is challenging to build NIR-capturing devices under the premise of hardware price reduction. Commercial NIR sensors are protected from modification. The process of building a new device is not trivial because it is required to start from scratch wit...
Contexts in source publication
Context 1
... to previous statements, the device is composed of the following modules: Processing unit, Cooling system, Capture Device, Illumination Printed Circuit Board (PCB), Display module, Power supply, and Case. All the modules are interconnected, as illustrated in the block diagram of Figure 1. ...
Context 2
... each resolution, we also performed IR tests using the method of [5], and scored the EER, as well as the d value. Figure 9 and Figure 10 show the relationship between the iris radius and the IR metrics. The elbow in Figure 9 occurs at an iris radius of 9.2px; however, EER stabilises at 46.9px. ...
Context 3
... elbow in Figure 9 occurs at an iris radius of 9.2px; however, EER stabilises at 46.9px. On Figure 10, the d also presents less rate of growth over 45px of iris radius. Figure 11 shows in more detail the IR tests at three different resolutions when the iris radius is 120.4 px, 46.9 px, and 9.8 px. ...
Context 4
... Figure 10, the d also presents less rate of growth over 45px of iris radius. Figure 11 shows in more detail the IR tests at three different resolutions when the iris radius is 120.4 px, 46.9 px, and 9.8 px. It can be seen that there is much more overlap at 9.2 px (Figure 11b) compared to the maximum resolution 120.4px ( Figure 11a); that is why d diminishes from 6.26 to 2.02. ...
Context 5
... 11 shows in more detail the IR tests at three different resolutions when the iris radius is 120.4 px, 46.9 px, and 9.8 px. It can be seen that there is much more overlap at 9.2 px (Figure 11b) compared to the maximum resolution 120.4px ( Figure 11a); that is why d diminishes from 6.26 to 2.02. The Detection Error Trade-off (DET) curves in Figure 11c reveal the system behaviour for the three resolutions. ...
Context 6
... 11 shows in more detail the IR tests at three different resolutions when the iris radius is 120.4 px, 46.9 px, and 9.8 px. It can be seen that there is much more overlap at 9.2 px (Figure 11b) compared to the maximum resolution 120.4px ( Figure 11a); that is why d diminishes from 6.26 to 2.02. The Detection Error Trade-off (DET) curves in Figure 11c reveal the system behaviour for the three resolutions. ...
Context 7
... can be seen that there is much more overlap at 9.2 px (Figure 11b) compared to the maximum resolution 120.4px ( Figure 11a); that is why d diminishes from 6.26 to 2.02. The Detection Error Trade-off (DET) curves in Figure 11c reveal the system behaviour for the three resolutions. It shows the expected result, that errors decrement with higher resolutions; however, at an iris radius of 46.9 px, the EER increases only by 1% with respect to the maximum resolution. ...
Context 8
... OpenCV [38] libraries used in the Raspberry-Pi program can only acquire images of 1,920 × 1,080 px from the cameras; however, the Jetson-Nano was able to read images of 3,264 × 1,848 px using Gstreamer. Figure 12 shows the relationship between the iris radius and the distance to the camera for each device. A red dashed line marks the 45px minimum iris radius found in the previous test. ...
Context 9
... to work with an iris radius over 45 px, the distance to the camera must not exceed 35 cm. According to Figure 12, the Raspberry-Pi gets a lower resolution with the camera v2 module than using the camera v1 module at the same distance to the camera. That is because the camera v1 module possesses a magnification lens, as shown in Figure 3a. ...
Context 10
... total, two capture sessions were taken, one during the day and one at night. This test can be seen in Figure 13. ...
Context 11
... means that at night the incident NIR light is not enough and the camera has to compensate for the gain, exposure and contrast at the expense of adding extra noise. Figure 14 illustrates how at night, there is much more camera noise present in the image. According to the day curve in Figure 13, over 40 cm of distance, the SNR starts to drop more rapidly. ...
Context 12
... 14 illustrates how at night, there is much more camera noise present in the image. According to the day curve in Figure 13, over 40 cm of distance, the SNR starts to drop more rapidly. Therefore, the maximum distance with good SNR is 40 cm, according to this test. ...
Context 13
... important aspect to take into account is that the users might not look straight into the camera, but instead be distracted by the screen. Despite the camera and the screen being physically close, the subjects' gaze is different when looking at the camera than at the screen, as illustrated in Figure 15. When looking at the camera, more iris surface is present in the image, whereas when the subject directs their gaze downwards, eye leads and eyelashes obstruct the iris. ...
Context 14
... the distances to the camera of 10, 15, 20, 25, 30, 40 and 50cm, an image was captured when the subject looked directly at the camera and another when looking at the screen. Then, the iris diameter along the x and y axis was measured for each eye (see Figure 15), and the mean ratio of Dy/Dx between the left and the right eye was scored for each frame. This ratio represents how much of the iris was occluded due to eye leads and eyelashes in the y axis with respect to the x axis, where the iris is never occluded. ...
Context 15
... ratio represents how much of the iris was occluded due to eye leads and eyelashes in the y axis with respect to the x axis, where the iris is never occluded. The results of this test are shown in Figure 16. When looking at the camera, the ratio settles at around 0.9 for distances over 25 cm. ...
Context 16
... obtain less eye-lead and eyelash obstructions due to the subject's gaze, the distance to the camera has to be above 25cm. Figure 17 illustrates those intervals. The intersection of the three produces the optimal distance to the camera that satisfies the three criteria, and it is the interval between 25 and 35cm. ...
Context 17
... DET curves shown in Figure 18 indicate that DenseNet10 has the best overall performance since its DET curve is lower than the others. CCNet has better performance than UNet xxs for FAR smaller than 15%. ...
Context 18
... If a 10% of False Match is permitted in the system, the three networks would have less than a 1% of False Rejections Rate. On the other hand, when the system allows only a 1% of FMR, UNet xxs doubles of FNMR of CCNet, at 4.35%, which is not as desirable. This corroborates that there is a trade-off between speed and performance when using Fig. 18: DET curves on ND-LG4000-LR ...
Similar publications
Iris Recognition (IR) is one of the market’s most reliable and accurate biometric systems. Today, it is challenging to build NIR-capturing devices under the premise of hardware price reduction. Commercial NIR sensors are protected from modification. The process of building a new device is not trivial because it is required to start from scratch wit...
