heathnaylor / analogous-fingerprints

Analogous Fingerprints (Research)

This is a research project to investigate and determine if a Convolutional Neural Network (CNN) can be used to associate never before seen fingerprints with known individuals that have only had a single fingerprint scanned before.

Dataset

This project plans to use the Sokoto Conventry Fingerprint Dataset to train the model.

Process

The dataset contains 6,000 fingerprints for 600 individuals

We will split the data as follows:

• x_train = 5,400 finger prints (One finger print omitted for each subject)

• Y_train = 5,400 labels (One finger print omitted for each subject)

  • These will be one-hot encoded for each of the 600 individuals

• x_test = 600 finger prints (the omitted one)

• Y_test = 600 labels (the omitted one)

  • These will be one-hot encoded for each of the 600 individuals

• The input layer will be the the raw image pixels passed through several convolutional, pooling, and dense layers

• The output layer will be a softmax layer with 600 outputs.