training a minutiae extraction model though a deep-learning network with ground truth latent friction ridge images as training samples,

wherein the deep-learning network comprises a base network configured to generate a minutiae feature map from a latent friction ridge image,

a Region Proposal Network (RPN) configured to propose minutiae locations and directions from the generated minutiae feature map at a given scale, and a Region-Based Convolutional Neural Network (RCNN) configured to fine-tune minutiae locations, and directions proposed by RPN,

wherein the training samples have at least marked minutiae positions and directions, and

wherein the training samples comprise images and marked minutiae of the same friction ridges each rotated by different angles;

preprocessing the latent friction ridge image by light balancing the image to produce a light-balanced latent friction ridge image;

inputting the latent friction ridge image and the light balanced latent friction ridge image into the minutiae extraction model to extract minutiae of the latent friction ridge image and the light balanced latent friction ridge image, wherein the model outputs sets of locations and directions for the extracted minutiae of each image; and

fusing both sets of the extracted minutiae using a majority voting system, wherein the majority voting system uses a modified non-max suppression algorithm, wherein if a majority of proximal minutiae have substantially the same direction then the remaining minority proximal minutiae are removed irrespective of the confidence score of the remaining minority proximal minutiae, to obtain a final minutiae data set.