Blur.me's deep learning facial recognition technology

Blur.me's deep learning facial recognition technology

Blur.me provides a convenient technology to alleviate your worries in an era where privacy is essential.
Today, we will take a look at the core technologies of Blur.me, face recognition, and blur technology.

How does Blur.me's facial recognition work?

Blur.me's facial recognition technology uses pre-trained deep learning models based on the latest artificial neural network baselines.
Let's take a look at the concept of face recognition using artificial neural networks.

First, the uploaded image is divided into grids of a predefined size. Then, each grid cell generates a bounding box ( bbox ) and predicts the confidence score for the existence of an object for each bounding box ( bbox ).
After that, a score is calculated by multiplying the object's existence probability distribution (class probability) by the cells and the reliability value of the bounding box. If the estimated value is lower than the threshold, the bounding box is removed, and the NMS (Non-maximum suppression) removes any duplicate boxes. After this process, the remaining bounding box ( bbox ) is the facial recognition data.

To wrap things up!

  1. Split the uploaded image into a grid.
  2. Each cell in the grid generates a bounding box (bbox) and predicts a confidence score for each bounding box.
  3. Create a probability distribution map ( class probability ) for each cell object.
  4. Calculate the score multiplied by the probability distribution for the object and the confidence of the bounding box.
  5. Removal of duplicate boxes by applying filtering and NMS based on the threshold set in the calculation result.
  6. The remaining bounding box ( bbox ) is the facial recognition data.

What is the concept of Blur.me's blurring effects?

Blurring is also known as smoothing, as it blurs things out of focus. It is used for various purposes, such as removing facial blemishes, smoothing out rough spots, and removing noise in photos.

As blurring is an effect used in our everyday lives, let's take a look at how the technology works!

On the original photo pixel, sampling is performed on a specified dimension. Then, a filter is applied to the sampled pixel information to obtain a new pixel value.

This process is repeated up to the last pixel (convolution operation), which creates a new blurred image as the final product.

Therefore, the result of the blur varies depending on the type of filter.

It's a relatively simple technique, but it can be bothersome if a photo is not smoothed out correctly, don't you think?

Today, we learned the concept of Blur.me's facial recognition and blur technology~!
I hope it's a bit easier to understand.

See you next time with more informative and interesting content~

Next : The concept of tracking technology for blurring faces out in videos !

A video is a sequence of multiple images played back in sequence.
In a video, all images of a selected person should be blurred out.
So, how can we determine that the face recognized in the current image is the same as the next image?
Next time, we will take a look at the tracking techniques for blurring out faces in a video.