Azure Image Analysis client library for Python

The Image Analysis service provides AI algorithms for processing images and returning information about their content. In a single service call, you can extract one or more visual features from the image simultaneously, including getting a caption for the image, extracting text shown in the image (OCR) and detecting objects. For more information on the service and the supported visual features, see Image Analysis overview, and the Concepts page.

Use the Image Analysis client library to:

• Authenticate against the service
• Set what features you would like to extract
• Upload an image for analysis, or send an image URL
• Get the analysis result

Product documentation | Samples | Vision Studio | API reference documentation | Package (Pypi) | SDK source code

Getting started

Prerequisites

• Python 3.8 or later installed, including pip.
• An Azure subscription.
• A Computer Vision resource in your Azure subscription.
  • You will need the key and endpoint from this resource to authenticate against the service.
  • Note that in order to run Image Analysis with the CAPTION or DENSE_CAPTIONS features, the Azure resource needs to be from a GPU-supported region. See the note here for a list of supported regions.

Install the Image Analysis package

pip install azure-ai-vision-imageanalysis

Create and authenticate the client

Using API key

To authenticate the ImageAnalysisClient using api key, you will need the endpoint and api key from your Azure Computer Vision resource in the Azure Portal. The code snippet below assumes these values are stored in environment variables:

• Set the environment variable VISION_ENDPOINT to the endpoint URL. It has the form https://your-resource-name.cognitiveservices.azure.com, where your-resource-name is your unique Azure Computer Vision resource name.

• Set the environment variable VISION_KEY to the key. The key is a 32-character Hexadecimal number.

Note: The client library does not directly read these environment variable at run time. The endpoint and key must be provided to the constructor of ImageAnalysisClient in your code. The code snippet below reads environment variables to promote the practice of not hard-coding secrets in your source code.

Once you define the environment variables, this Python code will create and authenticate a synchronous ImageAnalysisClient using key:

import os
from azure.ai.vision.imageanalysis import ImageAnalysisClient
from azure.ai.vision.imageanalysis.models import VisualFeatures
from azure.core.credentials import AzureKeyCredential

# Set the values of your computer vision endpoint and computer vision key
# as environment variables:
try:
    endpoint = os.environ["VISION_ENDPOINT"]
    key = os.environ["VISION_KEY"]
except KeyError:
    print("Missing environment variable 'VISION_ENDPOINT' or 'VISION_KEY'")
    print("Set them before running this sample.")
    exit()

# Create an Image Analysis client for synchronous operations,
# using API key authentication
client = ImageAnalysisClient(
    endpoint=endpoint,
    credential=AzureKeyCredential(key)
)

Using Entra ID

You can also authenticate ImageAnalysisClient with Entra ID using the Azure Identity library. To use the DefaultAzureCredential provider shown below, or other credential providers in this library, install the azure-identity package:

pip install azure.identity

Assuming you defined the environment variable VISION_ENDPOINT mentioned above, this Python code will create and authenticate a synchronous ImageAnalysisClient using Entra ID:

import os
from azure.ai.vision.imageanalysis import ImageAnalysisClient
from azure.ai.vision.imageanalysis.models import VisualFeatures
from azure.identity import DefaultAzureCredential

# Set the value of your computer vision endpoint as environment variable:
try:
    endpoint = os.environ["VISION_ENDPOINT"]
except KeyError:
    print("Missing environment variable 'VISION_ENDPOINT'.")
    print("Set it before running this sample.")
    exit()

# Create an Image Analysis client for synchronous operations,
# using Entra ID authentication
client = ImageAnalysisClient(
    endpoint=endpoint,
    credential=DefaultAzureCredential(exclude_interactive_browser_credential=False),
)

Creating an asynchronous client

A synchronous client supports synchronous analysis methods, meaning they will block until the service responds with analysis results. The code snippets below all use synchronous methods because it's easier for a getting-started guide. The SDK offers equivalent asynchronous APIs which are often preferred. To create an asynchronous client, do the following:

• Install the additional package aiohttp:

  pip install aiohttp
  

• Update the above code to import ImageAnalysisClient from the azure.ai.vision.imageanalysis.aio:

  from azure.ai.vision.imageanalysis.aio import ImageAnalysisClient
  

• If you are using Entra ID authentication with DefaultAzureCredential, update the above code to import DefaultAzureCredential from azure.identity.aio:

  from azure.identity.aio import DefaultAzureCredential
  

Key concepts

Visual features

Once you've initialized an ImageAnalysisClient, you need to select one or more visual features to analyze. The options are specified by the enum class VisualFeatures. The following features are supported:

1. VisualFeatures.CAPTION (Examples | Samples): Generate a human-readable sentence that describes the content of an image.
2. VisualFeatures.READ (Examples | Samples): Also known as Optical Character Recognition (OCR). Extract printed or handwritten text from images. Note: For extracting text from PDF, Office, and HTML documents and document images, use the Document Intelligence service with the Read model. This model is optimized for text-heavy digital and scanned documents with an asynchronous REST API that makes it easy to power your intelligent document processing scenarios. This service is separate from the Image Analysis service and has its own SDK.
3. VisualFeatures.DENSE_CAPTIONS (Samples): Dense Captions provides more details by generating one-sentence captions for up to 10 different regions in the image, including one for the whole image.
4. VisualFeatures.TAGS (Samples): Extract content tags for thousands of recognizable objects, living beings, scenery, and actions that appear in images.
5. VisualFeatures.OBJECTS (Samples): Object detection. This is similar to tagging, but focused on detecting physical objects in the image and returning their location.
6. VisualFeatures.SMART_CROPS (Samples): Used to find a representative sub-region of the image for thumbnail generation, with priority given to include faces.
7. VisualFeatures.PEOPLE (Samples): Detect people in the image and return their location.

For more information about these features, see Image Analysis overview, and the Concepts page.

Analyze from image buffer or URL

The ImageAnalysisClient has two overloads for the method analyze:

• Analyze an image from an input bytes object. The client will upload the image to the service as part of the REST request.
• Analyze an image from a publicly-accessible URL. The client will send the image URL to the service. The service will fetch the image.

The examples below show how to do both. The analyze from an input bytes object examples populate the bytes object by loading an image from a file on disk.

Supported image formats

Image Analysis works on images that meet the following requirements:

• The image must be presented in JPEG, PNG, GIF, BMP, WEBP, ICO, TIFF, or MPO format
• The file size of the image must be less than 20 megabytes (MB)
• The dimensions of the image must be greater than 50 x 50 pixels and less than 16,000 x 16,000 pixels

Examples

The following sections provide code snippets covering these common Image Analysis scenarios:

• Generate an image caption for an image file
• Generate an image caption for an image URL
• Extract text (OCR) from an image file
• Extract text (OCR) from an image URL

These snippets use the synchronous client from Create and authenticate the client.

See the Samples folder for fully working samples for all visual features, including asynchronous clients.

Generate an image caption for an image file

This example demonstrates how to generate a one-sentence caption for the image file sample.jpg using the ImageAnalysisClient. The synchronous (blocking) analyze method call returns an ImageAnalysisResult object with a caption property of type CaptionResult. It contains the generated caption and its confidence score in the range [0, 1]. By default the caption may contain gender terms such as "man", "woman", or "boy", "girl". You have the option to request gender-neutral terms such as "person" or "child" by setting gender_neutral_caption = True when calling analyze.

Notes:

• Caption is only available in some Azure regions. See Prerequisites.
• Caption is only supported in English at the moment.

# Load image to analyze into a 'bytes' object
with open("sample.jpg", "rb") as f:
    image_data = f.read()

# Get a caption for the image. This will be a synchronously (blocking) call.
result = client.analyze(
    image_data=image_data,
    visual_features=[VisualFeatures.CAPTION],
    gender_neutral_caption=True,  # Optional (default is False)
)

# Print caption results to the console
print("Image analysis results:")
print(" Caption:")
if result.caption is not None:
    print(f"   '{result.caption.text}', Confidence {result.caption.confidence:.4f}")

To generate captions for additional images, simply call analyze multiple times. You can use the same ImageAnalysisClient do to multiple analysis calls.

Generate an image caption for an image URL

This example is similar to the above, expect it calls the analyze method and provides a publicly accessible image URL instead of a file name.

# Get a caption for the image. This will be a synchronously (blocking) call.
result = client.analyze_from_url(
    image_url="https://aka.ms/azsdk/image-analysis/sample.jpg",
    visual_features=[VisualFeatures.CAPTION],
    gender_neutral_caption=True,  # Optional (default is False)
)

# Print caption results to the console
print("Image analysis results:")
print(" Caption:")
if result.caption is not None:
    print(f"   '{result.caption.text}', Confidence {result.caption.confidence:.4f}")

Extract text from an image file

This example demonstrates how to extract printed or hand-written text for the image file sample.jpg using the ImageAnalysisClient. The synchronous (blocking) analyze method call returns an ImageAnalysisResult object with a read property of type ReadResult. It includes a list of text lines and a bounding polygon surrounding each text line. For each line, it also returns a list of words in the text line and a bounding polygon surrounding each word.

# Load image to analyze into a 'bytes' object
with open("sample.jpg", "rb") as f:
    image_data = f.read()

# Extract text (OCR) from an image stream. This will be a synchronously (blocking) call.
result = client.analyze(
    image_data=image_data,
    visual_features=[VisualFeatures.READ]
)

# Print text (OCR) analysis results to the console
print("Image analysis results:")
print(" Read:")
if result.read is not None:
    for line in result.read.blocks[0].lines:
        print(f"   Line: '{line.text}', Bounding box {line.bounding_polygon}")
        for word in line.words:
            print(f"     Word: '{word.text}', Bounding polygon {word.bounding_polygon}, Confidence {word.confidence:.4f}")

To extract text for additional images, simply call analyze multiple times. You can use the same ImageAnalysisClient do to multiple analysis calls.

Note: For extracting text from PDF, Office, and HTML documents and document images, use the Document Intelligence service with the Read model. This model is optimized for text-heavy digital and scanned documents with an asynchronous REST API that makes it easy to power your intelligent document processing scenarios. This service is separate from the Image Analysis service and has its own SDK.

Extract text from an image URL

This example is similar to the above, expect it calls the analyze method and provides a publicly accessible image URL instead of a file name.

# Extract text (OCR) from an image stream. This will be a synchronously (blocking) call.
result = client.analyze_from_url(
    image_url="https://aka.ms/azsdk/image-analysis/sample.jpg",
    visual_features=[VisualFeatures.READ]
)

# Print text (OCR) analysis results to the console
print("Image analysis results:")
print(" Read:")
if result.read is not None:
    for line in result.read.blocks[0].lines:
        print(f"   Line: '{line.text}', Bounding box {line.bounding_polygon}")
        for word in line.words:
            print(f"     Word: '{word.text}', Bounding polygon {word.bounding_polygon}, Confidence {word.confidence:.4f}")

Troubleshooting

Exceptions

The analyze methods raise an HttpResponseError exception for a non-success HTTP status code response from the service. The exception's status_code will be the HTTP response status code. The exception's error.message contains a detailed message that will allow you to diagnose the issue:

try:
    result = client.analyze( ... )
except HttpResponseError as e:
    print(f"Status code: {e.status_code}")
    print(f"Reason: {e.reason}")
    print(f"Message: {e.error.message}")

For example, when you provide a wrong authentication key:

Status code: 401
Reason: PermissionDenied
Message: Access denied due to invalid subscription key or wrong API endpoint. Make sure to provide a valid key for an active subscription and use a correct regional API endpoint for your resource.

Or when you provide an image URL that does not exist or not accessible:

Status code: 400
Reason: Bad Request
Message: The provided image url is not accessible.

Logging

The client uses the standard Python logging library. The SDK logs HTTP request and response details, which may be useful in troubleshooting. To log to stdout, add the following:

import sys
import logging

# Acquire the logger for this client library. Use 'azure' to affect both
# 'azure.core` and `azure.ai.vision.imageanalysis' libraries.
logger = logging.getLogger("azure")

# Set the desired logging level. logging.INFO or logging.DEBUG are good options.
logger.setLevel(logging.INFO)

# Direct logging output to stdout (the default):
handler = logging.StreamHandler(stream=sys.stdout)
# Or direct logging output to a file:
# handler = logging.FileHandler(filename = 'sample.log')
logger.addHandler(handler)

# Optional: change the default logging format. Here we add a timestamp.
formatter = logging.Formatter("%(asctime)s:%(levelname)s:%(name)s:%(message)s")
handler.setFormatter(formatter)

By default logs redact the values of URL query strings, the values of some HTTP request and response headers (including Ocp-Apim-Subscription-Key which holds the key), and the request and response payloads. To create logs without redaction, set the method argument logging_enable = True when you create ImageAnalysisClient, or when you call analyze on the client.

# Create an Image Analysis client with none redacted log
client = ImageAnalysisClient(
    endpoint=endpoint,
    credential=AzureKeyCredential(key),
    logging_enable=True
)

None redacted logs are generated for log level logging.DEBUG only. Be sure to protect none redacted logs to avoid compromising security. For more information see Configure logging in the Azure libraries for Python

Next steps

• Have a look at the Samples folder, containing fully runnable Python code for Image Analysis (all visual features, synchronous and asynchronous clients, from image file or URL).

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.

When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information, see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.