首頁 > 網路資源 > 大同大學數位論文系統

Title page for etd-0829106-004632


URN etd-0829106-004632 Statistics This thesis had been viewed 1984 times. Download 13 times.
Author Shao-Wei Su
Author's Email Address No Public.
Department Communication Engineering
Year 2005 Semester 2
Degree Master Type of Document Master's Thesis
Language English Page Count 51
Title AUTOMATIC RED-EYE DETECTION AND CORRECTION BASED ON UNIFORM COLOR METRIC AND BINOCULAR GEOMETRIC CONSTRAINTS
Keyword
  • chroma
  • hue
  • Color difference
  • Uniform color space
  • Digital cameras
  • Digital photography
  • Red-eye
  • Red-eye
  • Digital photography
  • Digital cameras
  • Uniform color space
  • Color difference
  • hue
  • chroma
  • Abstract Red-eye is a highly objectionable defect that often occurs in digital images taken with a flash by modern small cameras. Although many red-eye reduction algorithms were proposed and equipped in most of the digital cameras, none of these algorithms is effective enough. In this thesis, an algorithm for automatic detection and correction of red-eyes is proposed. Multiple pairs of red-eyes snapped in different view angles are expected to be detected without human intervention, and red-eye colors are recovered to let the human eyes in photos have a more natural look. By representing the image in the uniform CIE Lab color space, the appropriate segmentation on three color axes is performed to extract the regions where red-eye color occurs. With the database containing all kinds of red-eye samples and the method of K-mean clustering, a more precise red-eye color filter is developed to sift out candidate red-eye regions. A similar color filter for skin tone is also developed. The correlation between the size of the human eyeball and binocular distance is employed to eliminate most false positives (image regions that look like red-eyes but are not). To further increase the accuracy, the shape of each red-eye candidate is inspected if it is a convex hull. The color of the detected red-eyes is finally corrected by modifying the saturation and luminance of the associated pixels such that red color is removed while maintaining a natural look. Test is performed on a data set of more than 200 photos with red-eyes of different sizes and tones. Simulation results show that more than 80% of red-eyes can be detected, 6% false positives, and colors of these detected red-eyes can be successfully corrected.
    Advisor Committee
  • Chun-hsien Chou - advisor
  • Chung-lin Huang - co-chair
  • Jia-Ching Jeng - co-chair
  • Files indicate in-campus access only
    Date of Defense 2006-07-04 Date of Submission 2006-08-29


    Browse | Search All Available ETDs