COMSATS Face: A Dataset of Face Images with Pose Variations, Its Design, and Aspects

Table 4

Proposed PAL approach.

---

Input: A set of input images with I = {1,2, ……., I} classes and J images of each class.

Do for i = 1, …, I

  (2) convert RGB images to grey,

  (3) estimate and crop face .

  (4) update mean and standard deviation of each image,

  (5) calculate mean image of each class, .

Final training images of each class, .

Initialize mislabelled distribution over m,

Do for

  (1) if t = 1, choose i samples per class for the learner.

  (2) train LDA feature extractor.

  (3) build a classifier h_t.

  (4) calculate pseudo loss, e_t

  (5) calculate

  (6) if , abort the loop

  (7) update the distribution

Final classifier of training image, hf .

Generate a matching score.

Output: Maximum matching score (M _score), I _recog = argmax (M_score).

---