Added:
-
Image Preprocessing
Contrast enhancement (adjustable 0.5-3.0)
Brightness adjustment (adjustable -0.5 to 0.5)
Sharpening filter with configurable strength -
Region of Interest (ROI) Detection
Focus OCR analysis on specific screen regions
Options: Full Frame, Bottom Third, Bottom Half, Center, Top Third
Reduces processing time and improves accuracy for specific layouts
Note: Not recommended unless uses for re-detecting videos that otherwise failed. -
Fuzzy Keyword Matching
Levenshtein distance algorithm for detecting misspelled/partial keywords
Configurable maximum edit distance (default: 2)
Catches OCR errors like "Director" → "Dlrector" -
Scrolling Credits Detection
Detects text position changes across consecutive frames
Identifies scrolling credits by tracking vertical movement
Configurable window size and position threshold -
Adaptive Frame Rate
Dynamically adjusts frame sampling based on text density
Automatically speeds up/slows down analysis -
Credit Structure Detection
Identifies role-name patterns (e.g., "Director: John Doe")
Recognizes structured credit layouts
Configurable role patterns and minimum role count -
Quick Preset Buttons
Best Quality: Optimized for accuracy
Best Speed: Optimized for performance
Fixed:
- Some hardware acceleration bugs.
- Confidence score now actually work, yay!
Known bugs:
- Intel QSV hardware acceleration not working, this will requires an updated to the ffmpeg built into Emby.
Please use VAAPI for now.
Please note:
Turning on all settings at once require a very large computer to have detections done within a reasonable time so set the settings you need for best results for you and leave the rest.