Shot-cut detection for B&W archive films using best-fitting kernel

A novel shot-cut detection method for black and white (B&W) archive films, employing the best-fitting kernel between two consecutive image frames is proposed in this paper. The best-fitting kernel, obtained using least squares, is used to estimate the second frame of each successive frame pair. The absolute frame difference between estimated and original image frames is utilized as criterion to decide a candidate scene-cut. A double thresholding approach, consisting of global and local adaptive thresholds, is used in this stage. Obtained candidates are forwarded to a confirmation stage to exclude single-frame defects, frequently encountered in archive film sequences. Experimental results show that the proposed method gives favorable result for B&W degraded archive films, compared to methods proposed in the literature.