Face Hallucination Using Bayesian Global Estimation and Local Basis Selection

Chih-Chung Hsu, Chia-Wen Lin

Department of Electrical Engineering
National Tsing Hua University
Hsinchu 30013, Taiwan

Abstract

This paper proposes a two-step prototype-face-based scheme of hallucinating the high-resolution detail of a low-resolution input face image. The proposed scheme is mainly composed of two steps: the global estimation step and the local facial-parts refinement step. In the global estimation step, the initial high-resolution face image is hallucinated via a linear combination of the global prototype faces with a coefficient vector. Instead of estimating coefficient vector in the high-dimensional raw image domain, we propose a maximum a posteriori (MAP) estimator to estimate the optimum set of coefficients in the low-dimensional coefficient domain. In the local refinement step, the facial parts (i.e., eyes, nose and mouth) are further refined using a basis selection method based on overcomplete nonnegative matrix factorization (ONMF). Experimental results demonstrate that the proposed method can achieve significant subjective and objective improvement over state-of-the-art face hallucination methods, especially when an input face does not belong to a person in the training data set.

Fig.1 The flowchart of the proposed method.

Key words: Face hallucination, manifold learning, super-resolution, prototype faces.

Database: https://sites.google.com/site/nthujesse/research/downloads

Source code (Matlab): https://sites.google.com/site/nthujesse/research/downloads

The source code including three algorithms are integrated in a single GUI. They are Park et al. 08, Yang et al. 08, and Freeman et al. 07's methods. We also integrate different basis decomposition methods such as PCA, NMF, and sparse NMF in it. Besides, the code also indulges different local refinement strategies such as global sparse coding or global sparse coding based refinements. For coeffieicnes caluclation, we provide four optimization solvers: Lasso, least-squares, newton's method, and gradient descent.

Experimental Results

*where Lasso algorithm is used in the facial-parts-only portion with ONMF bases and the global face image is hallucinated using the proposed method for justice.