A Hierarchical Framework for Facial Age Estimation

Algorithm 2

Algorithm flowchart of ranking-KNN.

Input: Training set 𝑆 with 𝑁 samples

   Testing sample 𝑥

   Group 𝑔 with boundary 𝛼_min and 𝛼_max, output by age group estimation

Output: Estimated age of the testing sample

Initialization: 𝑆⁻_𝑐=∅, 𝑆⁺_𝑐=∅, 𝑧_out=0, 𝑏=𝑎_max−𝑎_min𝑥, 𝑙(𝑐)=0

Discipline:

(1) For  𝑐=1∶𝑏

  (1.1) For  𝑖=1∶𝑁

       If  𝑦(𝑖)≤𝑐+𝑎_min−1  𝑥_𝑖∈𝑆⁻_𝑐

       Else 𝑥_𝑖∈𝑆⁺_𝑐

    End  for  𝑖

  (1.2) Call KNN to decide 𝑥 belonging to subset 𝑆⁻_𝑐 or 𝑆⁺_𝑐, using the samples within group 𝑔.

      If  𝑥∈𝑆⁻_𝑐, 𝑙(𝑐)=0  else  𝑙(𝑐)=1

  End for  𝑐

(2) Deduce the estimated age of testing sample based on 𝑙(𝑐):

    For  𝑐=1∶𝑏

    𝑧_out=𝑧_out+𝑙(𝑐)

    End for  𝑐

    𝑦=𝑧_out+𝛼_min

(3) Output the estimated age 𝑦