My check returned the same result thirteen times. It was measuring the scenery.

Описание и применение

Автоматический визуальный контроль видео может выдавать ложные положительные результаты, если измеряется неправильный регион. Новый подход, измеряющий субтитры каждую секунду, выявляет пропущенные субтитры и потенциальные проблемы с переполнением текста.

Задача

Automated visual checks can give false positives if they measure the wrong region or miss intermittent content, leading to undetected rendering failures.

Результат

The new check reliably detects missing subtitles and potential overflow, preventing silent failures that the previous constant‑margin check missed.

Рабочий процесс

• 1. Extract a single frame at the midpoint of each shot.
• 2. Compute pixel differences from the background to find ink bounds.
• 3. Measure margins (left, right, top, bottom).
• 4. Observe constant margins across all shots → false assumption of success.
• 5. Identify that margins are dominated by scenery (horizontal rule, progress bar).
• 6. Notice missing subtitles in one shot due to sampling during a pause.
• 7. Replace check: sample subtitle band once per second across entire duration.
• 8. Flag any gap >2.5 s as a warning.
• 9. Verify failure by pointing band at an empty region and confirming warning.
• 10. Adopt two rules: constant result over variable input is unreliable; a guard that never fails is invisible.

Доказательства

Thirteen identical lines, to the pixel.

Источник

The check could not fail.

Источник

It looks neither at the scenery nor at a single frame. It samples the subtitle band once per second across the whole duration.

Источник