Selecting the Best Frames from Videos for Palmprint Biometric Verification

In palmprint recognition, it is not always possible to obtain the desired quality image in the frames captured during image acquisition. Depending on the device or environment where the image is taken, problems such as shifting and blurring may occur in the images. In this study, a method is proposed to determine the best frames that can be used for palmprint recognition. This method detects the hand using the MediaPipe library and determines the quality of ROI regions using variance-based quality scores and manual labeling. Thus, the most suitable frame from which biometric data can be obtained can be determined. The proposed method achieved 99.15% accuracy in the training phase and 95.09% accuracy in the testing phase.