A high compatibility finger vein image quality assessment system based on deep learning. (15th June 2022)

• Record Type: Journal Article
• Authors: Ren, Hengyi; Sun, Lijuan; Guo, Jian; Han, Chong; Cao, Ying
• Is Part Of: Expert systems with applications. Volume 196(2022)
• Publication Date: 2022-06-15
• Subjects: Biometrics -- Finger vein -- Quality assessment -- Automatic labelling -- Deep learning -- Lightweight network; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.116603
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Finger vein feature recognition has been widely studied due to its high security and stability. However, the systematic errors caused by false and missing information in low-quality finger vein images are still concerns. Recently, researchers have proposed a finger vein image quality assessment scheme to eliminate low-quality images. The major drawback of this strategy is the dependence on limited domain knowledge as the strategy cannot effectively filter out low-quality images and is only adapted to a certain recognition method or database. In this paper, we design a new finger vein image quality evaluation method driven by improving the recognition performance of the recognition system and based on the possibility of the false rejection of low-quality images. The method uses statistical knowledge to analyse the matching of different samples between fingers so as to automatically determine the image quality and classify it. Then, we further design a lightweight convolutional neural network to train the high-quality and low-quality images obtained by the above evaluation criteria and mine the common attributes between low-quality images so as to ensure that the prediction system can judge the image quality quickly and accurately in practical use. The proposed scheme is validated on multiple public datasets and compared with current recognition algorithms. The experimental results show that the quality evaluation standard proposed in this paper has good universality and can be combined with different types of recognition methods and acquisition equipment to well screen out the existing low-quality images. Furthermore, the prediction method proposed in this paper has achieved excellent prediction performance on three public datasets, showing that the proposed method can effectively improve the performance of the original recognition system. Taking the local binary pattern (LBP) algorithm as an example, after using the prediction algorithm proposed in this paper to eliminate low-quality images, the recognition system improves the original recognition performance on the three datasets by 12.670% (SDUMLA), 11.923% (MMCBNU_6000) and 9.940% (FV-USM). Highlights: Novel finger vein image quality assessment system based on CNN. Novel finger vein image quality evaluation standard. Novel lightweight CNN to mine hidden difference attributes between low-quality images. Novel system is not limited to a certain recognition algorithm or data set.