Eigenface-domain super-resolution for face recognition

Abstract

Face images that are captured by surveillance cameras usually have a very low resolution, which significantly limits the performance of face recognition systems. In the past, super-resolution techniques have been proposed to increase the resolution by combining information from multiple images. These techniques use super-resolution as a preprocessing step to obtain a high-resolution image that is later passed to a face recognition system. Considering that most state-of-the-art face recognition systems use an initial dimensionality reduction method, we propose to transfer the super-resolution reconstruction from pixel domain to a lower dimensional face space. Such an approach has the advantage of a significant decrease in the computational complexity of the super-resolution reconstruction. The reconstruction algorithm no longer tries to obtain a visually improved high-quality image, but instead constructs the information required by the recognition system directly in the low dimensional domain without any unnecessary overhead. In addition, we show that face-space super-resolution is more robust to registration errors and noise than pixel-domain super-resolution because of the addition of model-based constraints.

People Identification and Tracking Through Fusion of Facial and Gait Features

“…There are two directions to handle the problem of low resolution. 1) super-resolution (SR) based methods [24], [38], [27], [28], [32], which reconstruct high-resolution images from low resolution images for visual enhancement. After applying SR, a higher resolution image can be obtained and used for recognition.…”

UHD Video Super-Resolution Using Low-Rank and Sparse Decomposition

“…The recovery of high-resolution (HR) images and videos from low-resolutions (LR) content is a topic of great interest in digital image processing with applications in many areas such as HDTV [11], medical imaging [20], satellite imaging [23], face recognition [12], immersive content generation, and surveillance [27]. The global super-resolution (SR) problem assumes that the LR image is a noisy, lowpass filtered, and downsampled version of the HR image.…”

A Novel Face Hallucination with an Error Regression Model and MPCA in RGB Color Space

“…Gunturk et al [8] apply PCA to determine the prior model. With learning-based method for face hallucination, a kernel PCA is applied for deriving prior knowledge about the face class [9].…”