Palmprint Recognition Based On Convolutional Neural Network

Palmprint recognition is one of the typical biological pattern recognition technologies,palm texture has uniqueness,stability and reliability,and palmprint image has large acquisition area,low requirement for acquisition equipment,and easy to be accepted by people,therefore,the research of palmprint recognition method has important theoretical and practical significance.As we all know,the extract of palmprint feature and the recognition algorithm are the key technologies in palmprint recognition,according to the problem of palmprint recognition under complex conditions such as image rotation and noise interference,combined with the algorithm advantage of Convolutional Neural Network(CNN),we propose an optimized convolutional neural network algorithm to extract palmprint features,and use Softmax classification method to classify the palmprints.The main research works are as follows:1)The establishment of CNN model is the key factor to determine the recognition rate,and it is also one of the difficulties in palmprint recognition using CNN.In this paper,the key parameters of CNN model are studied.According to the characteristics and experience of palmprint image,the method of establishing CNN model based on experiment is proposed.2)Due to the fact that there are few training samples for the authoritative palmprint database,and CNN training requires a large number of samples,at the same time,in order to improve the robustness of the recognition method to noise and rotation,Augmentation Processing of Image Samples method based on the spatial domain is proposed: The original palmprint database is expanded by adding Database With Noise and Rotate(DWNR).The model is established and tested in combination with the CNN modeling method proposed in part(1),it is proved that the proposed anti-noise and anti-rotation palmprint recognition method based on DWNR and CNN can recognize palmprint effectively,and has good robustness to noise and rotation.3)In order to improve the robustness of palmprint recognition system to noise and rotation,a large number of samples are added.Although it meets the requirements of CNN training and improves the performance of the proposed method,it increases the training complexity.In this paper,the Circularly Symmetrical Gabor Transforms(CSGT)method is used to decompose the image with low noise pollution,and we retain the other frequency components except the highest frequency component.By this method,on the one hand,it can improve the number of image samples,on the other hand,it plays a role in filtering noise with heavier noise.The number of samples is greatly reduced compared with method(2).Combined with the CNN modeling method in part(1),the palmprint recognition method based on CSGT and CNN is proposed.The experimental results show that the proposed method can effectively reduce the time complexity,improve the robustness and recognition rate of palmprint recognition compared with the palmprint recognition method in part(2).