Research on Inner Knuckle Pattern Recognition Method Based on Convolutional Neural Network

Abstract

In this paper, the hand inner knuckle pattern is the research object, the hand image is preprocessed by binarization, morphological processing, contour extraction, corner positioning, finger separation and knuckle ROI extraction. Then normalize the region of interest and form a two-dimensional matrix into the convolutional neural network for feature extraction. Finally we used the fully connected layer and the Softmax for classification and recognition. And studied the learning rate, the convolution kernels number, the neurons number in the fully connected layer, the convolutional layers number in the network and the impact of different optimization algorithms on the recognition results, obtain the best network parameters. Accroding to the experimental test and analysis, the recognition rate of the inner knuckle pattern recognition method based on convolutional neural network reached 95.2%, which has good application value.