How Can i improve the quality of documentai document-ocr processor result

Question

I have this image (first image) which I want to process using the document-ocr processor. The output I got looks something like the second image printed onto the python console. This output has been badly process in inclusion with most of my files. How can I make document-ocr understand and yield a perfect result.

Answer 1

To set expectations, no machine learning model can give "perfect results" consistently.

Results will greatly depend on the quality of the input files. In this case, the document is a scan of a handwritten file and handwriting can vary greatly from document to document.

In this particular example, some of these words could be difficult for humans to read, so the performance for Document AI could be inconsistent.

In general for improving quality of OCR output, higher quality for the source material results in higher accuracy of OCR detected text. So scanning at a higher DPI can improve results.

If the PDF file has embedded text already, then you can also use the Native PDF Parsing feature in the pretrained-ocr-v1.2-2022-11-10 processor version. This repository has some sample code for how to use it.

https://github.com/GoogleCloudPlatform/document-ai-samples/tree/main/pdf-embedded-text