Storage Space Reduction of Biometric Iris Databases by Successive Images Differences and Quadtree Decomposition

Abstract

In recent years, the rise of biometric applications, particularly those centered around iris-based systems, has been significant. High data volumes inherent in these applications and the potential vulnerability of network links necessitate data compression in certain instances. The advantage of lossless compression methods is twofold: they maintain recognition performance without degradation while necessitating fewer computations for differentiation compared to their lossy counterparts. This study proposes a novel approach for lossless/lossy compression of iris biometric sample data across various public iris databases. Initially, the differences between successive images within each class are calculated, leveraging the strong correlation of images within each class. Subsequently, these differences are compressed using quadtree decomposition. This methodology was tested on six renowned iris databases: CASIA V1, CASIA V3, MMU1, MMU2, and UBIRIS Iris, all of which contain 8-bit grayscale images. The results indicate that the proposed strategy offers superior compression performance across different iris databases in comparison to existing methods. Notably, the results suggest that this method can be effectively integrated into an iris biometric recognition system, providing efficient iris image compression, especially when applied in its lossless form.