Since facial expressions represent one of the most important ways in which humans shows their emotions and describe the human behavior, automatic facial expression recognition have received an important attention in the computer vision researching. Promising applications in machine human interaction and behavioral researches make facial expression recognition an interested topic.

Many problems concern this topic such as illumination changes, pose, angle of the input camera, partial occlusion, and so on. However partial occlusion is one of the most important problems in facial expression recognition because it can be seen as noise and disturb facial expression feature extraction or it will cause information loss, and also partial occlusion is very common in real life: glasses, mask, long hairs and face movements will cause different facial occlusions. Therefore, this work presents an analysis to determine the part of the face that contains the most important information and propose the use of `Sub-block Eigenphases' algorithm in the stage of feature extraction in order to develop an algorithm of robust facial expression recognition under facial occlusion conditions.

The proposed algorithm is divided in 3 main stages: facial region segmentation, features extraction and classification. In the stage of facial region segmentation, the face images are segmented into 4 face regions: forehead, eye-eyebrow, nose and mouth. Subsequently Sub-block Eigenphases algorithm is applied in the feature extraction stage, which uses the phase spectrum and PCA to extract the most important information of images. Finally Support Vector Machine (SVM), using the multi-class mode, is applied as a classifier to get the models and make the recognition of facial expressions.