Research On Contrast Enhancement And Blotch Detection & Removal Algorithms In Archived Film Restoration

films still have high artistic and cultural value. However, the quality of archived films has been affected by time and some unscientific measures to play and store films could have also caused various degradations such as contrast declination, blotch, scratch and flicker. This thesis investigates the problem of contrast declination and blotch artifacts in archived films. By taking advantages of the existed classic methods, novel algorithms of contrast enhancement and blotch detection & removal are both proposed, which aims to gain better restoration effect. The main innovations in this thesis are listed as follows:First of all, an example-based contrast enhancement algorithm is developed. In order to build a connection between the contrast of an example image and a target image, a histogram learning parameter is defined. Thus, contrast enhancement of a target image can be performed by learning from this parameter of an example image. When using this algorithm to restore the archived films, an adjacent context-similar frame with originally eye-pleasing or well-enhanced contrast can be utilized as an example image to enhance the target frame. Then, contrast continuity of successive frames can be satisfied in this way. Besides, this algorithm can also be applied for normal single image enhancement if the example image with similar context presents good visual quality.Second, on the basis of the Optimal Contrast-Tone Mapping (OCTM) algorithm, this thesis updates the objective function of contrast enhancement by re-defining image contrast, and adds some constraints to promise the optimal solution to be feasible, which improves the defects in the OCTM model like the unreasonable objective function and inadequate constraints. The improved optimal problem can also be solved by linear programming. In comparison, the proposed Improved OCTM algorithm has clearer definition toward the problem and more scientific and reasonable optimal model. Experimental results show this method is effective in contrast enhancement.Third, traditional blotch detection algorithms have relatively high false detection rate or computational complexity, while traditional blotch removal algorithms usually could not restore blotch edges very well. To settle the above problems, this thesis proposes a Non-Local Means based blotch detection & removal algorithm. Inspired by the non-local idea from the Non-Local Means algorithm, the blotch detection part extends the traditional SROD algorithm by making full use of the pixel information in the target frame and the adjacent frames. In the blotch removal part, Non-Local Means algorithm is utilized to restore the pixels in the detected mask. Experimental results show that the restored result is satisfactory especially in the blotch edges.