[Bug]: Auto Split PDF page dividers not recognised
Installation Method
Docker
The Problem
When using the Auto Split PDF function, and uploading a scanned document containing the QR code divider, the divider is not recognised.
Scan is 300 DPI (attached).
If I instead add the divider 'electronically' using the merge, organise, export functions, it works fine. Suggesting that it does not like something about the scanned QR code divider. I note that I have tried it using the newly released V2 website, and it splits fine.
Version of Stirling-PDF
1.5.0
Last Working Version of Stirling-PDF
No response
Page Where the Problem Occurred
No response
Docker Configuration
version: '3.3'
services:
stirling-pdf:
image: docker.stirlingpdf.com/stirlingtools/stirling-pdf:latest
ports:
- '7890:8080'
volumes:
- ./StirlingPDF/trainingData:/usr/share/tessdata # Required for extra OCR languages
- ./StirlingPDF/extraConfigs:/configs
- ./StirlingPDF/customFiles:/customFiles/
- ./StirlingPDF/logs:/logs/
- ./StirlingPDF/pipeline:/pipeline/
environment:
- DISABLE_ADDITIONAL_FEATURES=false
- LANGS=en_GB
Relevant Log Output
info.log shows the following when the Auto Split tool is trying to digest the scan:
2025-11-03 23:03:30,880 WARN o.a.p.p.g.c.PDICCBased [qtp1783192968-42] ICC profile is Perceptual, ignoring, treating as Display class
Additional Information
No response
Browsers Affected
Chrome
No Duplicate of the Issue
- [x] I have verified that there are no existing issues raised related to my problem.
Hi @canders1212! I'm Dosu and I’m helping the Stirling-PDF team.
The issue with scanned QR code dividers not being recognized in Stirling-PDF v1.5.0 is likely due to limitations in how that version processes scanned images—especially around image preprocessing, ICC profiles, and QR code detection. Electronically added dividers are cleaner and easier for the system to detect, while scanned dividers can be affected by scan quality, DPI, and color space handling. The ICC profile warning in your logs suggests the image rendering pipeline may not be optimally handling your scan, which can impact detection.
Recent updates in the newly released V2 (see PR #4446) introduced improved support for QR code divider sheets, duplex mode, and better image handling, which is why your scans work fine there. If possible, upgrading to V2 is the most reliable solution.
If you need to use v1.5.0, try rescanning with different settings (e.g., grayscale, higher contrast, or different DPI), and ensure your Docker image includes all necessary dependencies like OpenCV for image processing. Also, use the official divider sheet provided by Stirling-PDF (download link) for best results.
To reply, just mention @dosu.
How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
Really odd V2 website split it fine
Per my knowledge it should be running the same as v1.5 for backend code for this feature
I've just checked again and the v2 website handles it no problem.
I think we can rule out some kind of install issue on my end given that i'm using the stock docker-compose (other than port), and it works fine if I insert the divider page electronically rather than by scan...?
Really odd V2 website split it fine
Per my knowledge it should be running the same as v1.5 for backend code for this feature
I just updated my docker container and auto split stopped working. Using the files from here: https://github.com/Stirling-Tools/Stirling-PDF/issues/2281#issuecomment-2644132277
Pinning the image to 1.3.2-fat works, 1.4.0 and 1.5.0/latest don't work
edit: the pipline split stopped working, haven't tested the webinterface
Ahhh that explains it.. our V2 is based on a older version ATM we are in the works of updating V2 to pull latest OSS
Must be a bug introduced recently Let me investigate and resolve this thanks!