PHOW based feature detection for head pose estimation

In this paper, we use the Pyramid Histogram of Words (PHOW) feature in head pose estimation. Firstly, we divide the image into a regular grid, extract the Dense Scale-Invariant Feature Transform (Dense-SIFT) of it, and use K-means clustering method to obtain visual vocabulary dictionary. In order to save the location information of visual vocabularies, we set up the space pyramid of images and extract PHOW features based on Dense-SIFT. The method of feature extraction for head pose estimation is proved to be effective and is more robust to some noise (e.g., illumination, expression, occlusions and so on). Experimental results are presented to demonstrate the effectiveness of the feature extraction method.