Person Recognition Machine Learning Model

What is Machine Learning?

Machine learning is a branch of artificial intelligence (AI) and computer science which focuses on the use of data and algorithms to imitate the way that humans learn, gradually improving its accuracy.

The Goal of the Model

The goal of the model is to correctly identify people from faces from a given data set much faster than a human normally would be able to, with at least a 90% or greater accuracy from 4 or more different datasets.

Image A

Image B

Image C

Image D

The Data Collection Process

The data collection process involves using an algorithm to take over 250 images of a face in a short amount of time, repeated until there is a sufficiently diverse data set to compare between. Each image is then converted to grayscale, and downsized to a 48 by 48 image to increase model efficiency.

Processed Image A

Processed Image B

Processed Image C

Processed Image D

My Model

The model is an algorithm that sorts given images of faces into categories and identifies provided images of faces based on those categories. These images are converted into digital arrays of numbers that the algorithm compares to the other images, compares the given image to all data available to the algorithm, then identifies the image based on its conclusions.

My Data

Building the Model

The model is built using layers of Keras, with the main processing layer being made of 1024 neurons of Sigmoid activation, and the output layer having 4 layers of Softmax correction. The main processing layer uses 1024 cycles of processing to collect and compare data from a given image, and makes adjustments based on previous data and the new given data. The output layer has 4 layers to determine which category of face the main processing layer has determined. All together, the model has 7,083,012 parameters.

Accurately Labeled Images

Image A

Prediction: Matthew

True face: Matthew

Image B

Prediction: White

True face: White

Image C

Prediction: Kendrick

True face: Kendrick

These images are accurately labeled by the model, meaning that a random face was given from a dataset, and the model correctly identified which image it was shown.

Mislabeled Images

In any given machine learning model, it is bound to make mistakes or errors, given a sufficiently large and diverse dataset. With the limited dataset provided, however, the model had less data and parameters/details to process, thus increasing its accuracy.

Page updated

Report abuse