Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{198573,
        author = {Vallabh Khomane and Dr. Asmita Manna and Swarup Kusalkar and Kedar Kolambe},
        title = {Automated Visual Change Detection for Film Continuity Analysis using Training Data Derived from Composite Error Examples},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {11373-11379},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=198573},
        abstract = {Visual continuity errors are prevalent inconsisten- cies in film and video production that require meticulous manual checking by dedicated on-set and post-production personnel. This paper presents a deep learning approach to automate the detection of visual changes that may correspond to such errors. Motivated by the availability of composite images illus- trating known continuity mistakes—typically showing side-by- side frames with superimposed highlights—we propose a system that bypasses the need for manual pixel-level annotation. An automated preprocessing pipeline extracts corresponding image patches guided by visual highlights in the source composites and generates binary change masks based on thresholded pixel differences between the patches. A U-Net-like architecture, fea- turing shared encoder weights and difference-based skip con- nections utilising a custom AbsoluteValueLayer for robust serialisation, is trained on these automatically derived data triplets. The trained model produces a probability map indicating pixel-wise changes between two new input images. We detail the automated data-generation strategy, the network design, and evaluation results using standard segmentation metrics. We discuss the system’s effectiveness in localising significant pixel discrepancies and acknowledge the inherent limitations arising from the automated mask-generation process, positioning the tool as an aid for accelerating manual continuity review.},
        keywords = {Change Detection, Continuity Error, Deep Learning, U-Net, Siamese Network, Computer Vision, Video Analysis, Automated Preprocessing, Image Segmentation.},
        month = {April},
        }