In the LRCR, the input images are decomposed into two matrix: low-rank matrix and error matrix.
Finally, the algorithm of the LRCR is summarized in Table 1.
At last, we show the superior performance of the LRCR method for facial gender recognition under an unconstrained environment on comparison with 6 the state-of-art algorithms.
We compared the proposed LRCR method against other 7 state-of-the-art methods which including LBP+SVM, 2D-Gabor+SVM, LBP+Adaboost, 2D-Gabor+Adaboost, CRC without low-rank decomposition, DBN and CNN.
Primarily, it can be clearly seen that the proposed LRCR algorithm is remarkably superior to the other 6 methods in almost all possible input image size.
So the overall recognition performance of the CNN model is less effective than that of the LRCR, which shows that the model of deep multi-layer convolutional neural network is more suitable to solve the problem of many categories classification of big data like that does in the ILSVRC competition rather than the problem of binary-classification with relatively small image samples.
Moreover, we list the feature dimension of 5 features including LBP, Gabor, LRCR, DBN and CNN and the runtime of 7 facial gender recognition methods in Table 3 and Table 4, respectively.
All in all, compared with the other the-state-of-the-art methods, the LRCR method can effectively identify the gender from the nature face images captured in the unconstrained environment.