ESR 6: Making mobile biometrics more reliable
What has been achieved and the impacts
We presented the investigation on the effect of time of day on the matching accuracy of speaker recognition system. We have collected 1780 voice samples donated by 30 people. For evening versus evening comparison scenario EER of 1.46% was achieved. For the morning versus evening comparison scenario, the equal error rate is increased to 10.2%.
From a database of more than 1900 iris images from 509 eyes (723 diabetic iris images from 161 eyes and 1183 healthy iris images from 348 ones), we used three different matchers, (open source) and found that accuracy was consistently higher with those who do not have diabetes.
An equal error rate (EER) of 4.68% was achieved when identifying faces under the joint influences of full makeup and mood variation, while the EER under the effect of each of these factors separately is less than 1%.
Our study highlights the limits of unimodal biometric tech and cautions against the widespread use of modalities that only perform well in optimal circumstances and do not account for relatively common conditions.
The results of this project contribute to the contactless mobile biometric systems, which are important to the biometrics industry. Therefore, it will have a potential economic impact.
We also recommend the industrial partners that efforts should be undertaken in the biometric community to increase the reliability of security systems by combining existing unimodal biometric matches in order to implement multimodal authentication systems.
The biometric solutions provided and proposed in this research are widely applicable and can be easily adapted to the implemented application and environment. Improving the robustness of biometric systems can enhance the popularity of contactless mobile biometric systems. Robustness, which is defined as the survivability under failure or attack, is one of the most important properties of a system. A biometric system should perform well under any circumstances. In this work, we have detected several issues and fixed them by using the same methodology.