A Hierarchical Framework for Facial Age Estimation

Algorithm 1

Algorithm flowchart of SKNN.

Input: Train set with 𝑁 samples

   Test sample  𝑥

Output: age label 𝑦 of the testing sample

Initialization: 𝑧_𝑖=0, 𝐷(𝑖)=0(𝑖=1,2,…,𝑁), 𝑛_𝑐=0, 𝑐=1,2,…,𝐶

Discipline:

(1) Map age label into category label for all training samples

      𝑧_𝑖=𝑓(𝑦_𝑖𝑦)=floor^𝑖𝐿(𝑖=1,2,…,𝑁)𝑧_𝑖∈𝜔_𝑐

(2) Calculate the distance between test sample 𝑥 and all train samples

     𝐷(𝑖)=²(𝑥−𝑥_𝑖)∗(𝑥−𝑥_𝑖)^𝑇(𝑖=1,2,…,𝑁)

(3) Sort top 𝐾 smallest distances 𝐷(𝑘)(𝑘=1,2,…,𝐾) from 𝐷(𝑖)(𝑖=1,2,…,𝑁),

   then record their corresponding category label  𝑧_𝑘.

(4) Compute the sample number  𝑛_𝑐 belonging to different categories  𝜔_𝑐.

   𝑛_𝑐=sum(𝑧_𝑘=𝑐)  𝑐=1,2,…,𝐶

(5) Select the category label 𝜔_𝑘 with the maximum number  𝑛_𝑘, then map inversely the

   category label  𝜔_𝑘  label into age label 𝑦.

      𝜔_𝑘=armax(𝑛_𝑐)𝑦=𝑓⁻¹(𝜔_𝑘)=𝜔_𝑘×𝐿

(6) Output 𝑦