There have been lots of research about face detection and face recognition. However, faces with makeup usually seriously affect the face recognition result. If we want to recognize the face with higher accuracy, it would be better to first know whether the input face is with makeup or not, and we can use corresponding makeup face or non-makeup face model to recognize it. Unfortunately, the current available datasets for face analysis do not include enough makeup and non-makeup image pairs of users. In this work, we propose a framework to efficiently increase pairs of makeup and non-makeup face images for the existing makeup face datasets. Patch-based features are extracted and support vector machine (SVM) is applied to classify whether a face image is with makeup. The technique of partial least squares (PLS) is then employed to authenticate whether a makeup photo and a non-makeup photo belong to the same person. By combining the makeup detection and the face authentication methods, we can successfully construct a larger face dataset that can be specifically used for applications of makeup face analysis.