Abstract: In order to reduce noise and extract richer discriminant feature, the local ternary local derivative pattern operator was proposed. By determining whether the adjacent pixel gray values within a certain range, the local derivative pattern of current pixel was extended from binary to ternary, and the face feature vector was formed by sequentially connecting the uniform pattern of the space histogram of the encode. The chi-square statistic was used to calculate the sample similarity of face feature vector. The experimental results on ORL, Yale and CAS-PEAL-R1 face database show that the recognition performance of the proposed algorithm was superior to local binary pattern and local derivative pattern.