| dc.description.abstract | This study proposes a multimodal palm tattoo recognition method. This system captures palm images through a smartphone, using the method based on the largest inscribed circle to capture the area of interest palm print, and then uses the gray level co-occurrence matrix (GLCM), gray scale Gradient co-occurrence matrix (GGCM) and local binary pattern (LBP), three texture feature extraction methods to extract palmprint texture feature vectors respectively, and three sets of feature vectors are combined with a probabilistic neural network classifier (PNN) to obtain the three sets of best inferences probability. Finally, we weighted and fused the inference probabilities of each module to obtain the recognition result. The experimental results of the self-built palm image database show that with 10 training data, the accuracy of the three modalities of GLCM, GGCM, and LBP are 82%, 86%, and 89%, respectively. The accuracy was increased to 96%. We also use the public palm print database provided by Mutah University for comparative experiments, and the accuracy can reach 99.5%. Compared with VGG-19 and AlexNet, our multi-modal palm tattoo sub-recognition system has good recognition performance. | en_US |