Use Matlab

ROC curve plays an important role in binary classification problem. Here we take face verification as an example to learn how to draw a ROC curve. Labeled Faces in the Wild (http://vis-www.cs.umass.edu/lfw/) (LFW) dataset is widely applied in face verification benchmarks. We provide two facial descriptors in this question: LBP (https://en.wikipedia.org/wiki/Local_binary_patterns) and VGG-Face(http://www.robots.ox.ac.uk/~vgg/software/vgg_face/). The LBP feature is a 5900D vector while VGG-Face is a 4096D vector.

We have 6000 pairs of faces (in LFW_label.mat), and each row in this matrix indicates the face# of the two faces. The maximum # is 13233. Every 600 pairs will be one fold from the 10-fold-validation setting (Note, we dont need model training in this question, so we directly use every 600 pairs for testing). For every 600 pairs, the first 300 pairs are true face pairs, meaning the two faces are from the same person, and is assigned label 1, while the second 300 pairs are false pairs, meaning the two faces are from different persons, and is assigned label 0.

Please use all ten folds for testing.

Hint: If you would like to use built-in ROC curve function of MATLAB, there are two input parameters: targets and outputs. targets is the ground truth label, i.e., 1 or 0, while outputs is the similarity score you learned from the two faces you want to verify. Here I suggest to use cosine similarity for the outputs. In addition, if you think the dimensionality of LBP and VGG-Face features are high, you may want to use PCA first to reduce the dimensionality.

Your task is to show a ROC curve with false acceptance rate on X-axis and true acceptance rate on Y-axis for both LBP and VGG-Face features.

[provided files list]: LFW_LBP.mat, LFW_VGG.mat, LFW_label.mat

files are in this link https://drive.google.com/drive/folders/0B_W5TGttGNmzVXdQS001UTFsd0U