Impact of Image Quality on Face Detection
This project aims at investigating the interaction of perceptual image quality on computer vision tasks. WNCG Profs. Alan Bovik and Joydeep Ghosh, with student Suriya Gunasekar currently work on images with facial content and developed algorithms for face detection under commonly observed distortions in image transmission and storage including additive white noise, Gaussian blur, and JPEG compression.
The WNCG research team quantifies the degradation in performance of a popular and effective face detector when human–perceived image quality is degraded by distortions due to additive white gaussian noise, gaussian blur or JPEG compression. It is observed that, within a certain range of perceived image quality, a modest increase in image quality can drastically improve face detection performance. These results can be used to guide resource or bandwidth allocation in a communication/delivery system that is associated with face detection tasks.
A new face detector based on QualHOG features is also proposed that augments face-indicative HOG features with perceptual quality–aware spatial Natural Scene Statistics (NSS) features, yielding improved tolerance against image distortions. The new detector provides statistically significant improvements over a strong baseline on a large database of face images representing a wide range of distortions. To facilitate this study, the WNCG researchers created a new Distorted Face Database, containing face and non–face patches from images impaired by a variety of common distortion types and levels. This new dataset is available for download and further experimentation at www.ideal.ece.utexas.edu/˜suriya/DFD/.
Paper 1: Face Detection on Distorted Images using Perceptual Quality-Aware Features