Видео

There used to be a free trial where they give you some credit. Once that credit was used up, you would need to pay to use. I don't see that option on their website any longer. Currently they have a prepayment method, in $5 increments.

If you are using vscode as your editor, you can follow this video for general instructions on how to get these working together: ruclips.net/video/lGRwEcCHNtA/видео.html google-cloud-vision is the main package for connecting python to google cloud

You can, and Google Vision should work okay as a barcode reader. However, there are other tools that might be more useful and accurate as they were built specifically for this purpose. ML Kit and Android Google Mobile Vision Library are a few. I haven't used either of these.

Thanks for pointing this out - AWS made some changes to IPv4 costs in February of this year (2024).

Hi, textractor is a specialized package used only for connecting to the AWS Textract service, whereas boto3 is a general purpose library, it can be used for many different AWS API services. If you were just working with AWS Textract, the Textractor package would be easier to use.

creating a text document from a scanned book would be possible, but the accuracy may not be 100%. You would need to edit and correct the mistakes in the output after the OCR is finished. Creating the images would be more difficult.

@@TechExpertTutorials that's fine, but I'm looking a way to do it in one step, ocr and image recognition to a doc file, then, rearrange and correct it.

Hi, Google does not have an API designed for fingerprint recognition. You can check out the WebAuthn API, although I have never used it.

Thanks! To handle multiple images effectively, you could use the Azure OpenAI GPT4 service and include the BYOD "add your data" option, which can include extracting text from files/images. Then you can ask any questions about any of the files/images. The process would look through the data on all of the files and find the most relevant information. FYI, I plan to have a video out soon that will cover how to set this up.

yes, you can, i did it this way: content = [{"type": "text", "text": prompt}] for filename in os.listdir(images_folder_path): if filename.lower().endswith(('.png', '.jpg', '.jpeg', '.gif', '.bmp')): filepath = os.path.join(images_folder_path, filename) content.append({"type": "image_url", "image_url": {"url": f"data:image/png;base64,{encode_image(filepath)}"}}) response = client.chat.completions.create( model='gpt-4o', messages=[{"role": "user", "content": content}], max_tokens=500, )

i don't know if that's what you really need but i hope it works for you

You should be able to take the extracted text from Gemini AI and add that information to a web or text-based form. I don't know if there are any 3rd party tools that would simplify this, but it could be done.

I was able to use this with a free Google Cloud account previously. Google gives you $300 in free credits to use for testing and evaluation. Here are some more details: cloud.google.com/vision/docs/drag-and-drop

​@@TechExpertTutorials thanx 🤗

This is using the Google Gemini large language model version 1.5 to extract the text. This multimodal model includes multiple functions in one model. Find more details here: ai.google.dev/gemini-api/docs/text-generation?lang=python

@@TechExpertTutorials I was trying to do same task using open source models. The google ocr engine tesseract is terrible for such tasks, so I was trying to achieve it using the object detection models and then passing the bboxes to the ocr, but it's performance is not up to the mark. Any suggestion how to proceed with open source models?

@@BhrantoPathik I have found that open-source models are currently not as accurate as the paid models from OpenAI and Google Gemini. This could change in the future as larger and more robust models are built and made available in open-source. I also tested pytesseract, see this video for details: ruclips.net/video/UBpFPBVlINw/видео.html

You need billing enabled so if you go above 1,000 calls/units it can bill you accordingly

thanks!

thank you!

Hi, thanks for watching and subscribing. I am still working on the next video for this topic. Also, to ensure that only logged in users can access this would require setting up a login form and a database on the website, to store the credentials. Microsoft Azure does provide additional ways to secure access, such as firewalls, gateways, key vaults, etc.

Thanks! I would suggest Azure, it has the best accuracy, and can even be improved upon if you include an OpenAI ChatGPT component.

Not sure what is the accuracy for other languages, I have only used the English version. Here is a list of other languages that are available for this API: cloud.google.com/vision/docs/languages There is a paper comparing different languages, including Arabic, and it mentions the accuracy for Arabic is lower than English (93% vs 98%): www.researchgate.net/publication/356446235_OCR_with_Tesseract_Amazon_Textract_and_Google_Document_AI_a_benchmarking_experiment

Hi, thanks! AWS Textract does not currently read and decode QR Codes. You might be able to do that using a different AWS Service. There is an open-source library named ZBar that can decode QR codes and Bar codes.

Thanks for letting me know. The repo has been moved. Here is the new location: github.com/TechExpertTutorials/AzureAIVision this is also updated in the description

@@TechExpertTutorials