This research article examines image quality and the use of the Feature Information Measure (FIM) to improve the accuracy of identification.
The authors content that poor image quality negatively affects iris recognition accuracy and feature information is too objective of a measure to evaluate the iris image quality alone. The authors propose that combining Feature Information Measure (FIM), an occlusion measure, and a dilation measure, a quality score can be obtained that correlates with positive recognition accuracy. In the author’s model, FIM is calculated as the distance between the distribution of iris features and a uniform distribution, allowing for even low contrast images performing well in iris recognition. To adjust for low contrast image/low FIM scores, the author’s developed an information-based contrast invariant iris quality measure and performed comparison in CASIA 1.0, CASIA 2.0, ICE, and West Virginia University databases, as well as applied the convolution matrix, spectrum energy, and Mexican hat wavelet approaches. The author’s experimental results show that the proposed quality measure can predict matching performance. (Publisher abstract provided)