Algorithm 1: The Proposed H-CRC Algorithm.

1. input: A dictionary that contains training set and a test sample $\mathbf{y}$. 2. preprocessing: Perform the pose normalization using a 3D morphable model. 3. fori = 1 to N (the number of test samples) do 4. Calculate the histogram residual of the test sample for each training sample by Equation (12). 5. Choose the best L histogram measurements with the smallest Euclidean distances for the test sample by Equation (13). 6. Represent the test sample using a linear combination of the obtained L nearest neighbors by Equation (14). 7. Evaluate the sum of the contributions of achieved neighbours across each class for representing the test image by Equation (16). 8. Perform the classification that produces the smallest reconstruction error by Equation (17). End for Return: The final recognition rate.