The invention relates to an anti-cheat detection method for a human face in an identity authentication system, which comprises the steps of firstly, extracting spatial information of pixels by using a local binary pattern, grayscale distribution statistics and grayscale co-occurrence matrix to obtain texture features of a space domain; secondly, extracting a low frequency complex coefficient and a high frequency complex coefficient by using two-dimensional dual-tree complex wavelet decomposition to obtain texture feature of a frequency domain; then performing feature fusion by using PCA dimension reduction so as to fuse the texture features of the space domain and the texture features of the frequency domain; and finally, performing feature fusion on the texture features of the space domain and the texture features of the frequency domain, and detecting and judging a real/fake human face image by using an SVM classifier. According to the invention, the texture features of the space domain and the texture features of the frequency domain are fused, especially the texture features are extracted by using the time shift invariance and the direction selectivity of two-dimensional dual-tree complex wavelet decomposition in the frequency domain, and dimension reduction and decorrelation are performed on the fused features by using PCA, so that the calculation complexity is low, the redundancy is low, the consumption of time and space is saved, the accuracy of human face cheat detection is improved, and the security of human face cheating in the identity authentication system is enhanced.