Since automatic face recognition in the wild still suffers from low-quality, low resolution, noisy, and occluded input images that can severely impact identification accuracy, this article presents a novel technique that improves the quality of such low-resolution face images beyond the current state of the art.
The project modeled the correlation between high and low resolution faces in a multi-resolution pyramid and recovered the original structure of an un-seen extreme low-resolution face image. By exploiting domain knowledge of the structure of the input signal and using sparse recovery optimization algorithms, the project recovered a consistent sparse representation of the extreme low-resolution signal. The proposed super-resolution method is robust to noise and face alignment, and it can handle extreme low-resolution faces up to 16x magnification factor with just seven pixels between the eyes. Moreover, the formulation of the proposed algorithm enables simultaneous occlusion removal capability, a desirable property that other super-resolution algorithms do not possess. Most importantly, the project shows that the demonstrated method generalizes on real-world low-quality surveillance images. (publisher abstract modified)