A quality-based context-switching framework for palmprint recognition

Palmprint is one of the most important biometric modalities used for human identity recognition due to its significant number of discriminative features. In the literature on palmprint recognition, there is a wide variety of methods categorized into three main groups: feature-based, holistic-based, and hybrid methods. Feature-based and holistic-based methods experience a decline in recognition accuracy due to variations in the quality of biometric samples. In an effort to enhance recognition accuracy, researchers have introduced hybrid methods. However, these methods are more complex in terms of computational cost. Our paper aims to introduce a novel hybrid approach called a quality-based context-switching framework for palmprint recognition to address both recognition accuracy and computational cost limitations. This framework comprises three components: quality assessment, context-switching, and palmprint recognition. The first component is a quality assessment algorithm that calculates a set of image quality metrics. These metrics are then utilized by the second component (context-switching algorithm), which is implemented by a Support Vector Machine (SVM) to dynamically select the appropriate palmprint recognition method. Based on the classification of state-of-the-art palmprint recognition methods, the final component of our framework may consist of any two methods chosen from the local feature-based and holistic-based categories, respectively. Experimental results on the Hong Kong Polytechnic University palmprint database version 2 (PolyU-II) demonstrate that the proposed framework achieves the best tradeoff between recognition accuracy and computational cost compared to the methods proposed in the literature.