How structured video content breaks scrolling screenshot stitching on Android

A developer maintaining an Android app that stitches scrolling screens into long images discovered it produced overlapping captures when pointed at social media feeds containing videos. Investigation revealed that playing videos create a coherent secondary motion signal that misleads pixel-matching algorithms, causing rows of content to be placed at incorrect positions in the final image. A common fix—excluding moving rows from measurement—backfired because it relied on the very offset it was meant to verify, causing the mask to discard valid page data and protect the erroneous video signal instead. The developer resolved the issue by splitting the visible frame into horizontal slices and using consensus across slices to determine the true scroll offset, without anchoring to any prior estimate. This anchor-free consensus approach ensures that when agreement is insufficient, the frame is rejected outright rather than committed with a wrong position.