We're sorry Aspose doesn't work properply without JavaScript enabled.

Free Support Forum - aspose.com

Aspose Sample OCR Code in Github Yielding Extremely Poor Accuracy

Aspose.OCR Product Family
#1

I downloaded sample code from the Aspose Github repo: https://github.com/aspose-ocr/Aspose.OCR-for-Java/tree/master/Examples.

The first problem is that the code does not compile against the version of the product specified in the Maven POMs. So, I fixed the code to compile.

The second problem is that when I run PerformOCROnImage against sample1.jpg, also provided by Aspose in the repo, the results are horrible, less than 5% is accurate.

We are very interested in purchasing this product if we could get it working to some reasonable extent. We are running it on the lastest Mac OS X and JDK. Perhaps, someone could help?

#2

I tested the same sample code on a Windows machine and got the same results, with accuracy worse than 5% correct.

#3

@juane3729,

The sample input file that you have used is for demonstration of “Recognition Block” example. Please use the attached sample input files for evaluation.
samples.zip (188.5 KB)

#5

Thanks for your reply. sample1.bmp works perfectly. However, sample3.png has extremely poor accuracy (< %5 correct). sample3.png doesn’t seem like it should be that hard to read. We’ve also had extremely poor accuracy feeding in some of our own files.

Is there is something else I am missing. Could it be related to fonts?

#6

@juane3729,

Fonts could be an issue as Aspose.OCR APIs currently support Arial, Times New Roman, Courier New, Tahoma, Calibri and Verdana in Regular, Bold and Italic styles. Please forward us sample input file(s) that you are using at your end. We will test it and update you about our findings.

#7

As I mentioned, one of the files that didn’t work was sample3.png, which you attached earlier inside the zip.

#8

@juane3729,

We are able to observe the issue that the text returned by OCR operation has many discrepancies in it. The issue has been logged into our system with ID OCRJAVA-771 for further investigation by our Product team. We’ll update you here once there is some information or a fix version available in this regard. We are sorry for the inconvenience caused.