FKQNet: A biometrie sample quality estimation network using transfer learning

It is worth mentioning that the use of image quality assessment can play a significant role to achieve the performance improvement for biometric recognition frameworks. The aforementioned works and other analogue studies in low-resolution finger knuckle images have claimed that the precise detection of true features is difficult from poor quality images and the main reason for matching errors. The quality of finger knuckle images are mainly affected by blurred lines, random skin folds, unclear wrinkles, de-focus, poor contrast and varying reflections produced by the camera flash. Due to the lack of well-structured patterns as in the case of face or fingerprint biometrics, the quality assessment of knuckle images becomes challenging. In this paper, we have proposed a novel quality Network (FKQNet), that classifies FKP Images based on six different quality parameters. To the best of our knowledge this is the first attempt, a trained Deep Learning Neural Network has been employed to identify, estimate and quantify the quality attributes of knuckle images. Following this, an image has been classified into three classes viz. good, bad and average and on that basis, the best samples are selected or low weights are assigned to poor quality samples for further level recognition. The objective of the proposed system is to enhance the matching performance of biometric recognition frameworks through the use of quality assessment of image samples. The experimental results obtained from the publicly available finger knuckle database reveals that the proposed method is highly competitive compared with other state-of-the-art approaches.