Dear Nikita ,
I will try your notes, and I appreciate your prompt attention to this matter
Regards, Osama
Dear Nikita
When i scan the attached document there’s some bubble has been marked with lightly shade with a pencil. but the result is not correct as it recognized as empty bubble
I tried to adjust the threshold to 20 but it recognized all the bubble even the empty as filled when i increases the threshold to 25 some lightly shade bubble it recognized as empty
Is there any option to solve this problem ?
Hello, @osa8am82
Can you please share an example of template scan, that was marked with light shade of pencil?
I will investigate the issue with this example.
Hello, @osa8am82
I have investigated provided example.
Yes, some bubbles, like questions from 24 up to 34, are ignored by the recognition engine. They are not bright enough to be considered a valid answer by OMR engine.
Lowering threshold in GlobalPageSettings or in specific elements, unfortunately, will not help.
I can recommend using only strong and bright bubble marks, as in 35-45 questions.
Otherwise, they could be considered as a mark removed by eraser or accidental smearing of bubbles nearby etc.
Dear Nikita
Most of all student’s they use pencil to fill up the bubbles & all OMR apps I’ve used like “Remark Office” can detect the light shade of pencil except Aspose.OMR, i think this issue must be considered with your developers and mark it as major technical defect
Regards, Osama
@osa8am82 I am sorry to hear that. I will investigate for possible solutions and get back with development time frames.
1 Like
I have been working on an OMR-based student answer sheet correction project for the past five months and have chosen Aspose.OMR as the core component for the correction process. However, I have encountered a critical issue that may prevent me from continuing to use your product.
The problem lies in the detection of lightly shaded pencil marks. The current implementation fails to recognize answers where students have shaded the bubbles lightly, which significantly impacts the accuracy of the correction process. Since standardized OMR sheets often include various shading intensities, this limitation poses a major challenge for our application.
I would appreciate it if your development team could address this issue soon. If there are any workarounds or configuration settings that could improve the recognition of light pencil marks, please let me know. Additionally, if a fix is planned for an upcoming release, I would appreciate an estimated timeline.
I am happy to provide sample images to help reproduce the issue if needed. Looking forward to your prompt response.
Regards, Osama
Hello, @osa8am82
Thank you for your patience,
I have investigated the described issue with detection of lightly shaded pencil marks.
We can offer a solution in a form of a new calibration mechanism. To properly recognize lightly shaded pencil marks, it will be required to fill by hand a calibration example page.
Results of this calibration can be used later on in Recognize() method against real scans. Approximate workload will be about 2-3 weeks.
I will consult with colleagues to approximate release date.
Dear Nikita,
Thank you for your prompt response and for investigating the issue.
We appreciate your efforts in proposing a calibration-based solution to improve the recognition of lightly shaded pencil marks. This approach sounds promising, and we are open to testing it once the calibration mechanism becomes available.
Kindly keep us updated regarding the estimated release date, and please let us know if there are any preliminary steps or documentation we can review in the meantime.
Also, please confirm if you would like us to prepare and send a sample page filled by hand with pencil marks for calibration purposes.
As an additional suggestion, we have noticed that software like Remark Office OMR offers a visual indication of problematic areas—specifically highlighting the circles with marking issues—allowing users to easily identify and review them during the recognition process. Implementing a similar feature in your solution would significantly enhance its usability and accuracy.
In addition, it would be a great improvement if the recognition engine could automatically adapt to different paper sizes regardless of the scan resolution (in terms of pixel density). This would make the system more flexible and robust across various input sources.
Looking forward to your updates.
remark.jpg (54.3 ك.ب)
Best regards,
Thanks for sharing your feedback and suggestions.
Yes, we will surely inform you and keep you updated with the status of the investigation. We will surely include the information in the process that has been shared by you and as soon as we have some definite updates in this regard, we will let you. Please be patient and spare us some time.
PS: Yes, it will be appreciated if you can prepare and send a sample page filled by hand with pencil marks.
Dear Asad,
Thank you for your response.
Just to clarify, I have already shared a sample filled page in one of my previous comments.
However, during the upcoming week, I will be conducting a practical and detailed test at a private school using my own project, which I have developed based on Aspose.OMR.
The test will involve 40 students across 9 different exams, and I believe this real-world scenario will provide valuable feedback and help evaluate the accuracy and reliability of the solution.
I’ll be happy to share the results with you once the testing is complete.
Best regards,
Hello, @osa8am82
Thank you for your patience!
I have added Calibration feature in the nearest future of our roadmap and will publish it in a 25.5 release, in the May 2025.
Dear Nikita,
I apologize for my late response, as I was on leave during the past period.
Regarding the future update, I believe it will be an excellent enhancement to Aspose.OMR, as it will improve the credibility and accuracy of the results.
As I mentioned earlier, we conducted the exams recently and instructed students to mark their answers with dark shading. Initially, I encountered an issue where multiple choices were being recognized for some questions—even if they weren’t actually marked. However, after adjusting the threshold value to 27, the recognition became significantly more accurate.
this issue raises a concern, we are using various types of scanners across different schools, and having to manually adjust the threshold for each device would be a major limitation. It would not be practical to fine-tune the threshold setting for every scanner used.
With the threshold set to 27, the output has been very reliable. I’ve also added a new feature to my application that displays both the correct answer and the student’s marked choice directly on the answer sheet. This makes it easy for parents to review their child’s responses and understand any grading discrepancies. So far, the results have been promising.
Still, I remain concerned about the variation in scanner hardware and whether Aspose.OMR can consistently handle different scan qualities—especially when detecting pencil shading. I’ve attached a sample exam for your reference.
Note, many schools here use an app called “ZipGrade”, which relies on a mobile phone camera to detect marked answers very quickly. This eliminates the dependency on scanner quality and could be worth exploring as a complementary approach.
Thank you once again for all your efforts. I truly appreciate your support and hope we can continue working together to make Aspose.OMR one of the best OMR solutions available.
Regards, Osama
Corrected.jpg (380.7 ك.ب)
Screenshot 2025-05-02 122230.png (367.0 ك.ب)
Screenshot 2025-05-02 122316.png (301.7 ك.ب)
Hello, @osa8am82
Thank you for your feedback!
I will investigate on how to make the result of the calibration process more universal, to support different quality of scanning.
Regarding mobile phone camera, they could be a great tool, but they would require one to hover the phone over each sheet, which limits the number of sheets that could be processed. Quality of photo is usually good, especially in latest smartphones, but poor lighting could create a false positive.
Aspose.OMR at the current moment targets recognition of images produced by automatic scanners, which is especially easy with the help of Batch feature. There are potential issues with scanner artifacts, but they more manageable and predictable.
But I will note it for future research, thank you!
Hello, @osa8am82
Thank you for your patience! Release for pencils is almost ready and will be published soon!
Hello, @osa8am82
Light pencil shade detection is ready and published in 25.6 release of Aspose OMR for .NET.
Boolean flag for additional processing is available at
Aspose.OMR.BatchProcessings.BatchTemplateProcessor.ApplyLightShadeProcessing
Would appreciate your feedback on quality of recognition.
Thank you for your patience!