In this paper, an Enhanced Total Variation based Quotient Image model (ETVQI) is described and discussed in detail. In the ETVQI model, a histogram equalization method is adopted to enhance contrast of samples. Then the TV-L1 model is chosen to decompose samples to large-scale part u and small-scale part v. With the large-scale part, samples are further normalized by dividing them by u to normalize signals of intrinsic structures. Lastly, some feature fusion methods are adopted to generate the final normalized sample as the result of ETVQI. To apply ETVQI to face recognition, subspace analysis algorithms are suggested to perform subspace analysis on normalized samples. According to experiments on the CAS-PEAL face database, the face samples preprocessed by ETVQI could improve the performance of some famous subspace analysis algorithms (PCA, KPCA and ICA) with the standard testing sets proposed in the face database. Experimental results also confirm that samples preprocessed by ETVQI could make PCA, KPCA and ICA robust to not only lighting, but facial expression, masking, occlusion etc. in face recognition area.