| Title | Fast Retinex Method Based on CMSB-plane for Variable Lighting Face Recognition |
| Author | *Su Young Lee, Seok Lai Park, Young Kyung Park, Joong Kyu Kim (Sungkyunkwan University, Republic of Korea) |
| Page | pp. 729 - 732 |
| Keyword | Retinex, adaptive smoothing, CMSB-plane |
| Abstract | In this paper, we propose a fast Retinex method based on CMSB-plane for variable lighting face recognition. Cast shadow, created by direct light sources, seriously damage face images and ultimately deteriorate the performance of face recognition system. To eliminate cast shadow efficiently in the framework of Retinex theory, they should be preserved in the illumination estimation process. The proposed method estimates the illumination with cast shadow by iteratively convolving the input image with a 3×3 averaging mask adaptively weighted by coefficients based on both the combined most significant bit-plane (CMSB-plane) and image entropy. To improve the performance of the smoothing process, the proposed method employs the multigrid method and combines multi-copies of the estimated illumination with various scales. In this way, we can achieve a fast illumination normalization in which even face images with cast shadow are normalized efficiently. The proposed method has been evaluated based on the CMU PIE database by using PCA. In result, the proposed method has higher recognition rates than other conventional illumination normalization methods such as SSR, MSR and SQI. |
| Title | Multi-view Range Image Registration using CUDA |
| Author | *Sung-In Choi, Soon-Yong Park (Kyungpook National University, Republic of Korea), Jun Kim, Yong-Woon Park (Agency for Defense Development, Republic of Korea) |
| Page | pp. 733 - 736 |
| Keyword | GPU, CUDA, Multi-view, Range, Registration |
| Abstract | In this paper, we propose a real-time and on-line 3D registration system which acquires and registers multi-view range images simultaneously. The proposed system implements a 3D registration technique using GPU programming techniques. To register multi-view range images accurately in real-time, the repetitive parts of projection and transformation, which require a large computational overhead, are processed using CUDA. Using a portable range sensor with 320x240 resolution, the system shows about 5 frames per second of matching performance. |
| Title | Edge-Sensing Demosaicing based on Modified Edge Indicator |
| Author | *Hyun Mook Oh, Moon Gi Kang (Institute of TMS Information Technology, Yonsei University, Republic of Korea) |
| Page | pp. 737 - 740 |
| Keyword | Color interpolaton, Bayer pattern, Directional interpolation, Cross-channel correlation |
| Abstract | In this paper, we propose a demosaicing algorithm which interpolates the missing pixels on the estimated edge direction with modified edge indicator. We use gradients to estimate the edge direction which considers cross-channel correlation in addition to the within-channel correlation. After the edge direction is estimated, we interpolate the missing color values more precisely in aid of weighted color difference values. The experimental results show that the proposed method outperforms the conventional methods, both on objective and subjective criteria. |