Solution 1 :
I suspect the image resolution is the issue here. It’s that or you are hitting the limit of what the detection algorithm is capable of.
Could you try CameraX’s MlKitAnalyzer API? It automatically configures MLKit to give you the best result.
I’m using Firebase Vision, from MLKit, together with CameraX for text recognition in Android. The problem is that the results are not very consistent. Depending on the font, these recognitions may vary. For example, sometimes it’s confusing the character ‘b’ with the digit ‘6’. This is not a problem if you only need digits in your result. But sometimes it’s confusing digit ‘7’ with digit ‘1’, which is quite annoying for a digit detector.
Is there any way to improve these results? Larger bitmaps could help? Any ideas on what can I do?