Project: VisionOCR
Codename: function_5
Method: Cloud
Lang/Dev Env: C#, .NET 9 - WinForms - Visual Studio 2022.

A lightweight Windows desktop OCR utility built with C# + WinForms (.NET 9) that batch-processes images using Azure AI Vision (Image Analysis) - Read (OCR).

This started as an AI-102 study side-project: I had screenshots and image “flash cards” where copy/paste wasn’t possible, and I wanted a faster way to convert them into clean text notes.

This project helped me understand how to integrate Azure's Vision services into a desktop application. To do that, I had to whip up a WinForms application to test the OCR capabilities. I'd do WPF, but I don't want to spend too much time on UI/UX design. WinForms is quick and easy for prototyping. WPF would've taken me three days to get something working. WinForms took me a couple of hours. Even that diverted me from study!

Azure Cognitive Vision services was much easier_ than AI search and Language/Speech services. My last project involved many: STT, TTS, Language Understanding, and AI Search. That project (STT - function0) is located here: https://github.com/sorrynofocus/STT

• Developers learning Azure AI Vision OCR
• AI-102 exam preparation
• Quick desktop OCR utilities for screenshots and images

This project is a Windows Forms application for batch OCR processing using Azure Cognitive Services - Vision (Image Analysis).

I used the ImageAnalysisClient class to extract text from images using the READ feature. The raw output is then processed to clean up text and formatting issues. In this application example, I focused on images containing text, such as screenshots, labels, and signs. There's no support for PDFs, and Word documents. That's for Document Intelligence services that support such formats and layouts. The samples live in the img_samples sub folder.

• Drag and drop multiple images
• Sends each image to Azure AI Vision OCR (Read)
• Extracts raw OCR lines
• Runs a small cleanup pipeline:
  • Removes common “junk UI lines” (configurable list in code)
  • Normalizes OCR whitespace (NBSP, mixed newlines, extra spaces)
  • Attempts basic paragraph reconstruction
• Displays the result in a RichTextBox
• “Save All” outputs text files per image

Supported image formats in this example:

JPEG, PNG, BMP, TIFF

The Cognitive Vision OCR engine is meant for photos, screenshots, labels, street signs, UI captures, etc.

A quick demonstration how the example application works.

In this image exhibit you can see the meter text was extracted, along with text. No AI responsibility was included in this example.

In this image exhibit you can see the text from various signs. The Vision service reads in natural reading order (left to right, top to bottom), thus giving a mixed output. Unless specifically bounded in regions, which we could, it is not implemented in this example.

In this image exhibit, the meme photo is quite large. The Vision service was able to extract all text. In this example, I did not find the social handle it is referring to, but Vision Ai can pick up email addresses, URLs, phone numbers, hash tags, and social handles. It may have hallucinated it. But, it did pick up the rest of the text quite well.

Note: OCR confidence and entity detection may vary depending on image quality. The service may occasionally misinterpret or hallucinate symbols or handles.

Should you build this project locally, ensure you have the following prerequisites:

• Windows 11
• Visual Studio 2022 or .NET SDK installed
• Azure subscription
• An Azure AI Vision / Computer Vision resource (Image Analysis / OCR)

NuGet dependency:

• Azure.AI.Vision.ImageAnalysis (1.0.0)

  NOTE: There is a auto-build.cmd script included to help automate the build process. Just run it in the project root folder.

• README.md
  Project overview, setup instructions, and workflow explanation.

• img/
  README screenshots and workflow diagrams.

• img_samples/
  Sample images used for OCR testing.

• ocr_function5/
  WinForms application source code.

  • VisionOCRClient.cs - Azure Vision OCR wrapper
  • TextProcessProvider.cs - OCR cleanup + paragraph reconstruction
  • auto-build.cmd - Publish helper

• scripts/
  Azure setup and teardown automation.

  • ocr_function_5_runbook_azure_setup.md - READ this first for the services setup/teardown runbook
  • 000_boot_env.cmd - Global environment bootstrap
  • setup/ - Provision Azure resources
  • teardown/ - Remove Azure resources

1. VisionOcrClient.ExtractRawLinesAsync() calls Azure Vision AnalyzeAsync(..., VisualFeatures.Read)

2. Raw line text is extracted from result.Read.Blocks[].Lines[]

• TextProcessProvider applies:

• RemoveDropPhraseLines(...)

• NormalizeOCRText(...)

3. ConstructPotentialParagraphs(...)

4. Output is shown in the UI and saved to text files

Legend: End User = person using the application (you?)

• Use the provided setup scripts to create the required Azure resources.

• This includes the Azure AI Vision (Image Analysis) service, which provides the OCR capability.

• The end user enters Endpoint and API Key into the application.

• These credentials authenticate requests to Azure AI Vision.

NOTE: The Endpoint must be the resource endpoint, for example: https://my-vision-resource.cognitiveservices.azure.com/ Not the regional API root.

• When processing begins, each selected image is uploaded temporarily to Azure’s OCR endpoint.

• Azure does not store the images; they are processed in memory as part of the request.

• Azure’s Image Analysis “Read” feature extracts text lines from the image.

The service returns structured OCR results containing:

• Detected text lines
• Blocks
• Bounding boxes (not used in this app, or in debug mode view)
• Confidence scores (only viewable in debug mode!)

The raw OCR lines are cleaned using your custom pipeline:

• Removes keyphrases (hard-coded in application. See TextProcessProvider.DefaultDropPhrases)
• Normalize whitespace, newlines, tab, non-breaking spaces.
• Reconstruct paragraphs

Note: Insane comments are included in source. I'm a huge commenter as most are not. I feel good comments are for readability and comprehension. if (annoyed) {sorry = true}

• The cleaned OCR text is displayed in the UI.
• End User can save all extracted text to files for later use.

In my last project, I gave a really long explanation of how to create the many services. I ended up creating scripts for automation. Luckily, there is a couple of services we need to create. So, I created scripts to automate the process.

If you need to modify the scripts, feel free to do so. The scripts are located in the scripts folder. The only script you really need to modify is the boot loader -- that sets all environment variables for the scripts. Once that's done, all you need to do is pass in your subscription ID OR set it as an environment variable AZ_SUBSCRIPTION. That easy!

This project uses a Service Principal to authenticate automation scripts that deploy and tear down Azure resources.

NOTE: Refer to the document ocr_function_5_runbook_azure_setup.md in the scripts folder for detailed setup and teardown instructions.

Based on Azure SDK for .NET samples (for ImageAnalysisResult):
https://github.com/Azure/azure-sdk-for-net/blob/main/sdk/vision/Azure.AI.Vision.ImageAnalysis/README.md
https://github.com/Azure/azure-sdk-for-net/blob/main/sdk/vision/Azure.AI.Vision.ImageAnalysis/README.md#extract-text-from-an-image-file
https://learn.microsoft.com/en-us/cpp/cognitive-services/vision/imageanalysis-imageanalysisresult
https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/how-to/call-read-api

Cognitive Services - Vision (Image Analysis) documentation:
https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/overview
https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/overview-image-analysis
https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/concept-ocr
https://github.com/Azure/azure-sdk-for-net/tree/main/sdk/vision/Azure.AI.Vision.ImageAnalysis

Other issues like detecting list items, paragraph reconstruction, and whitespace normalization are referenced in below articles:
https://learn.microsoft.com/en-us/azure/cognitive-services/computer-vision/overview-ocr
Non-Breaking Space (U+00A0) Unicode Character
https://unicode-explorer.com/c/00A0
Trim characters
https://learn.microsoft.com/en-us/dotnet/standard/io/file-path-formats#trim-characters

Why Stop at Words? Unveiling the Bigger Picture through Line-Level OCR
https://arxiv.org/html/2508.21693v1

How OCR Works — Line Segmentation
https://how-ocr-works.com/OCR/line-segmentation.html

Sentence Boundary Detection in OCR Text
https://aclanthology.org/W19-4006.pdf

GCP Vision OCR Line Segmentation Example
https://github.com/7codeRO/line-segmentation-gpc-vision-ocr

Finally...
https://regex101.com/ is your friend!

This document was assist-generated by CoPilot/Chat GiGPT for readability and ease of use.