It is often difficult to deal with the problem of 2D-face recognition under unconstrained conditions. The objective of this study is to develop an original method that overcomes such obstructions. The proposed approach combines a holistic method, the Principal Component Analysis (PCA) to a local method, the Steerable Pyramid (SP). All tests were run on IV2 database, with challenging variability and including 3500 to 5000 comparisons by experiment from 315 different people. The followed protocol was established in the first evaluation campaign on 2D-face images using the multimodal IV2 database. Comparison with five submitted algorithms as PCA, LDA and LDA/Gabor provides satisfying results.