fbpx

What is computer vision ?

Share This Post

Share on facebook
Share on linkedin
Share on twitter
Share on email

Computer vision is a task achieved with mathematics and computer science to get insight from images or videos. The basic concept is to calculate some descriptors of a given image to analyze it and extract valuable information.

10 years ago, these tasks were performed with a programmatical approach resulting in a long processing time meaning that real-time processing was really challenging.

But now Deep Learning for Computer vision has arrived! We can now achieve complex analyses of images in record time thanks to an artificial intelligence approach.


But how does a computer see?

Now, most of the time computer vision is based on Deep Learning (DL), a field of Artificial intelligence.

It relies on a network of artificial neurons (convolutional neural networks), inspired by the human brain. It’s a superposition of layers with different goals and different levels of information. Each one of them will have a specific role in order to achieve a certain goal.

Like a human brain, some neuronal layers will be in charge of extracting high-level info like object localization, shape, or size. Some others will be in charge of extracting low-level info like texture.

Then the layers that we call the “head” of the neural network will be in charge of the recognition task based on all the previously extracted information.


What can an AI do concretely ?


         Segmentation:  partitions an image into multiple regions or pieces to be examined separately.

  • Object detection:  identifies specific objects in an image in order to perform different activities like counting, tracking, trigger some action like stopping a car, etc.

  • Facial recognition: One can teach an AI to recognize some specific person in order to perform identity validation for example.

  • Pattern detection: Is the process of recognizing repeated shapes, colors or other visual indicators in images like texture ( wood, iron, grass for example )

  • Image classification: Group images into different categories.

  • Feature matching: Detect similarities between pictures to smartly filter your pictures or find some objects matching your search for example.



How can I build an AI?

In order for the computer vision algorithm to be able to recognize an
image, it is necessary to train the neural network beforehand. To do
this, it is provided with a visual database, that has first been
annotated manually depending on the type of information it wants to
extract.

    1) Collect the Data
    2) Prepare your Dataset and annotate it
    3) Train an AI models
    4) Chose the best model
    5) Put it in production    


As you can see, building an AI is an iterative process but it always begins with annotation.

It’s really time-consuming and needs to be done carefully because it will be the key factor of success and performance for your AI.

 

You might also like

Get product updates straight in your inbox!

ipsum velit, eleifend eget fringilla ut libero dolor. mattis nunc venenatis amet,