For problem of bespectacled face recognition, a non-uniform local Gabor binary pattern or method is proposed. The method utilizes local Gabor binary pattern to represent face image, so as to extract multi-scale and multi-directional feature information of the image and to be able to be insensitive to light and noise robustness in identifying. Also, non-uniform zoning policy is presented, and the policy can also discriminate contents information while ensures adequate access to image-space information, to weaken role of glasses, etc. to the disadvantage of discrimination of interference information. RNLGBP method makes full use of the characteristics of random subspace method and NLGBP method, complementing each other's shortcomings while maximizing their strengths, in order to maximize recognition performance, and experimental results show that the bespectacled face recognition rate of this method is improved significantly compared to traditional methods.