M.Sc. Tezi Görüntüleme

Student: Bekir DİZDAROĞLU
Supervisor: Asst. Prof. Dr. Ali GANGAL
Department: Electrical and Electronics Engineering
Institution: Graduate School of Natural and Applied Sciences
University: Karadeniz Technical University, Turkey
Title of the Thesis: A Hybrid Approach Model for Detection and Restoration of Defects in Old Color Films
Level: M.Sc.
Acceptance Date: 4/12/2006
Number of Pages: 168
Registration Number: di576

      Old films are subject to degrade in quality due to bad environmental factors and repeated projection. Dust and dirt are major defects. They adhere to the film surface and appear as blotches. The blotches occur randomly in each frame and do not generally occupy the same spatial location in successive frames. Vertical scratches occur in a frame when the film is abraded by dirt particles in the projector. They occupy the almost full height of the frame and occur in nearly the same spatial place in several frames. Various other defects occur due to water damage or excessive heat.

      Digital image sequence restoration techniques are generally classified in three steps: motion estimation, defect detection and restoration of damaged locations. Accurate motion estimation and compensation usually are necessary for detection and correction of detected damages. The detection of which pixels are likely to be damaged is required for filling in only those areas containing damaged pixels. Lastly, the missing areas are filled in by restoration methods.

      In this thesis, an automatic restoration algorithm is presented for detection and concealment of local defects such as Gaussian noise and blotches in old color films. Firstly, in the proposed method, Gaussian noise is removed by spatiotemporal non-local means algorithm. Secondly, after predictive diamond search process is a fast motion estimation method, which is required for an accurate defect detection especially and restoration of damaged regions, the defects are detected SDI method which can be simply adapted to detection of defects in color image sequences. Finally, spatiotemporal inpainting methods based on the existing exemplar-based inpainting are used for correction of the damaged regions that are detected. True motion estimation is not required for filling in the missing regions in the proposed restoration methods. The results indicate that the proposed method restores successfully the damaged regions in general.

      Key Words:

Blotch, Defect, Inpainting, Motion Estimation, Predictive Diamond Search, Noise Removal, Missing Region, Filling in, Restoration