Disclosed are an offline identity authentication method and an apparatus which are used to solve the problem that the prior art relies on a human face database provided by the Ministry of Public Security and that it is difficult to identify a three-way consistency of a chip avatar, a document surface avatar and a certificate holder image. The method comprises: acquiring two or more images for identify authentication to construct a to-be-authenticated multivariate image group (101); extracting a cascaded PCA convolution feature for each image in the to-be-authenticated multivariate image group to obtain a corresponding eigenvector (102); merging the image information of the to-be-authenticated multivariate image group according to the eigenvector and a fractional fusion strategy with a supervised signal to obtain a corresponding merged vector (103); and sending the merged vector to an SVM classifier which has been preliminarily trained to determine the consistency of the image in the to-be-authenticated multivariate image group, so as to obtain an identity authentication result (104).