Blotch Inpainting Based on Sift Feature Extraction in Old Films

Old films may be subject to various defects, such as blotches, scratches, noise, flicker, fading and color changes. Here, one of the most common defects is the blotches. They are characterized in containing approximately constant intensity and irregular shapes and also in a random spatial position in successive frames. In this paper, a spatiotemporal inpainting technique based on the feature extraction in the blotch regions is proposed to not consider dependence on the motion estimation. First, based on the presence of blotches in the current video frame, the scale-invariant feature transform (SIFT) is used for the feature extraction in the proposed method. Then, the method implements a spatiotemporal algorithm for correction of blotch regions, considering the feature extraction points. The results are compared with the partial differential equations (PDEs)-based inpainting model, which is frequently used to fill small gaps in the image, and the video inpainting method. © 2019 Chamber of Turkish Electrical Engineers.