This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Super-Resolution for High Magnification Face Images Yi Yao 1 , Besma Abidi 1 , Nathan D. Kalka 2 , Natalia Schmid 2 , Mongi Abidi 1 The University of Tennessee, Knoxville, Tennessee, 37996 West Virginia University, Morgantown, West Virginia, 26506 [email protected] ABSTRACT Most existing face recognition algorithms require face images with a minimum resolution. Meanwhile, the rapidly emerging need for near-ground long range surveillance calls for a migration in face recognition from close-up distances to long distances and accordingly from low and constant resolution to high and adjustable resolution. With limited optical zoom capability restricted by the system hardware configuration, super-resolution (SR) provides a promising solution with no additional hardware requirements. In this paper, a brief review of existing SR algorithms is conducted and their capability of improving face recognition rates (FRR) for long range face images is studied. Algorithms applicable to real-time scenarios are implemented and their performances in terms of FRR are examined using the IRIS- LRHM face database . Our experimental results show that SR followed by appropriate enhancement, such as wavelet based processing, is able to achieve comparable FRR when equivalent optical zoom is employed. Keywords : Super-resolution, face recognition, high magnification, wide area surveillance 1. INTRODUCTION Visible facial features are critical for successful face recognition, which imposes a minimum requirement of resolution on the corresponding image acquisition systems [22, 23]. For instance, a resolution of 60 pixels between eyes is recommended by a well-know recognition engine FaceIt ® developed by Identix . Although PTZ cameras are commonly used in indoor and outdoor surveillance systems, most of them provide a maximum optical magnification in the range of 20×~30× . From our experiments, this amount of magnification can cover an area with a radius of 20m while maintaining the aforementioned resolution. For long range (>50m) surveillance and target identification, this magnification level is insufficient. There exist hardware and software solutions for an improved resolution. The hardware approach needs additional equipment such as high zoom lenses and/or high resolution CCD’s. Either choice results in exponentially increased cost. In comparison, software based approach relies purely on image post-processing and gains improved resolution from temporal relation of consecutive frames. Super-resolution is one exemplary software based method and has been receiving intensive research interests since the pioneer work of Tsai et al.  and abundant SR algorithms have been proposed [5, 6, 13-15, 19-21]. In , Borman et al. reviewed existing SR algorithms and divided them into two categories: frequency domain based and spatial domain based methods. A more detailed and updated review can be found in . In addition to general SR methods, algorithms designed particularly for face images also exist in literature, such as hallucinating faces  and Eigenfaces . designed particularly for face images also exist in literature, such as hallucinating faces  and Eigenfaces ....
View Full Document
This note was uploaded on 04/22/2010 for the course MI IP taught by Professor Vladbalan during the Spring '10 term at Universidad del Rosario.
- Spring '10
- The Land