A two-step approach to hallucinating faces: global parametric model and local nonparametric model

By Ce Liu, Harry Shum, Changshui Zhang. Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001, 2001

384 citations recorded by the source corpus.

In this paper, we study face hallucination, or synthesizing a high-resolution face image from low-resolution input, with the help of a large collection of high-resolution face images. We develop a two-step statistical modeling approach that integrates both a global parametric model and a local nonparametric model. First, we derive a global linear model to learn the relationship between the high-resolution face images and their smoothed and down-sampled lower resolution ones. Second, the residual between an original high-resolution image and the reconstructed high-resolution image by a learned linear model is modeled by a patch-based nonparametric Markov network, to capture the high-frequency content of faces. By integrating both global and local models, we can generate photorealistic face images. Our approach is demonstrated by extensive experiments with high-quality hallucinated faces.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/397f725d9695f6d7e3ca9dfdf0992971cbe36743

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Hallucinating face by eigentransformation (2005, methodology): “[4] developed a two-step statistical modeling approach integrating global and local parameter models.”
Multilinear Discriminant Analysis for Face Recognition (2007, background): “SUBSPACE learning is an important direction in computer vision research [3], [9], [12], [13], [29].”
Super resolution using edge prior and single image detail synthesis (2010, methodology): “Learning based techniques estimate high frequency details from a large training set of HR images that encode the relationship between HR and LR images (e.g., [10, 19 , 9, 2, 24, 5, 29, 18]).”
Modeling object appearance using Context-Conditioned Component Analysis (2015, methodology): “Combining Parametric and Non-Parametric Models for Appearance Synthesis Liu et al . combined a parametric subspace model and a local non-parameteric model in a two step procedure to solve the problem of face hal-lucination [17, 18].”
A Comprehensive Survey to Face Hallucination (2013, methodology): “Although the method (Liu et al. 2001, 2007a) is introduced in Sect.”
Ultra-Resolving Face Images by Discriminative Generative Networks (2016, background): “In order to generate high-resolution (HR) face images from low-resolution (LR) inputs, face hallucination [1,2,3,4,5,6,7,8,9,10,11,12] attracted great interest in the past.”
Face Hallucination: Theory and Practice (2007, methodology): “In this paper we focus on facial appearance modeling as in [24], but address additional practical considerations.”
Image hallucination with primal sketch priors (2003, background): “This is, in spirit, similar to face hallucination [3] and other related low-level learning work [4] [5] [7].”
Hallucinating face by position-patch (2010, methodology): “The approach to generate a high-resolution image from a single low-resolution image, with a set of training images from the same scene is called single-frame super-resolution [2,7–14, 25, 30 ].”
Super-resolution through neighbor embedding (2004, methodology): “Some methods make use of a training set of images [1, 9, 10, 11, 12, 14, 19], while others do not require a training set but reThis is the Pre-Published Version”