The use of transfer learning technique in recognition of people identity using palmprint on the first multimodal dataset including palmprint and ear images

Many algorithms have been proposed for recognition of palmprint in the past, most of which are based on the extracting features from the domain transformation. Many of these domains feature conversion are not rotate-invariant, therefore a large amount of preprocessing is used to align images. Manual selection of features is often time consuming and subjective, so prior knowledge is needed. Consequently, the choice of convolutional property through artificial intelligence methods and CNN structure will be effective and some of the disadvantages of extracting traditional features will be resolved. Through CNN layered structure, deep learning can yield good outcomes in self-learning features. In this paper, a complete system of authentication through palmprint lines based on transfer learning and Alexnet network is designed, which results in accuracy of ۹۳ precision on collected dataset and ۹۵.۹ % on IITD dataset. Our dataset is the first multi-modal data, consisting of ۳۶۰۰ ear and palmprint images of ۱۸۰ people.

Keywords:

palmprint , CNN , mutli-modal.