https://github.com/czesctuklap/emotion-recognition-project

The project develops a facial emotion classifier using the k-Nearest Neighbors (kNN) algorithm. The classifier uses Histogram of Oriented Gradients (HOG) and Principal Component Analysis (PCA) for dimensionality reduction with usage of normalisaton, preprocessing and augmentation.
https://github.com/czesctuklap/emotion-recognition-project

emotion-detection emotion-recognition face-recognition facial-expression-recognition fer knn

Last synced: about 1 year ago
JSON representation

The project develops a facial emotion classifier using the k-Nearest Neighbors (kNN) algorithm. The classifier uses Histogram of Oriented Gradients (HOG) and Principal Component Analysis (PCA) for dimensionality reduction with usage of normalisaton, preprocessing and augmentation.

• Host: GitHub
• URL: https://github.com/czesctuklap/emotion-recognition-project
• Owner: czesctuklap
• Created: 2024-05-16T19:35:41.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2024-11-24T18:23:37.000Z (over 1 year ago)
• Last Synced: 2025-02-03T14:39:45.702Z (over 1 year ago)
• Topics: emotion-detection, emotion-recognition, face-recognition, facial-expression-recognition, fer, knn
• Language: Jupyter Notebook
• Homepage:
• Size: 1.24 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Emotion Recognition Classification

The project is based on the `kNN` algorithm, which classifies objects based on the class of their nearest neighbors. It involves calculating the distances between points, selecting the nearest k distances, and adopting the feature that dominates in the k set. We used the Euclidean distance for distance calculation, which for x_i and x_j in n-dimensional space is defined as:

![euclidean](https://github.com/czesctuklap/emotion-recognition-project/assets/164773624/22b849af-c412-4204-ba14-3aee65012e6f)

where:

- \( x_i = (x_{i,1}, x_{i,2}, ..., x_{i,n}) \) is the feature vector for point i,

- \( x_j = (x_{j,1}, x_{j,2}, ..., x_{j,n}) \) is the feature vector for point j.

However, before the `kNN` algorithm is invoked, the data is normalized using the `StandardScaler()` function, labeled using the `LabelEncoder()`, and the dimensions are reduced using the `PCA` function.

Further in the code, the `preprocess-image()` function is used to process images before their use in the machine learning model. In this part, image normalization is also performed by subtracting the mean pixel value and dividing by the standard deviation:

![normalization](https://github.com/czesctuklap/emotion-recognition-project/assets/164773624/ee40d459-5985-454f-8432-caca0fd86a57)

where:

- `mean(img)` denotes the mean pixel value of the image,

- `std(img)` denotes the standard deviation of the pixel values of the image.

Next, the training data is augmented using the `ImageDataGenerator()` function, which applies various transformations to the images, such as rotations, shifts, zooms, and horizontal flips. Additionally, a function utilizing Histograms of Oriented Gradients (HOG) is defined, which extracts HOG features from both training and testing images. This sequence of steps prepares the data for machine learning, ensuring its diversity through augmentation and extracting important features using HOG, which can improve the model's ability to generalize and classification effectiveness.

As a result of the program's execution, we obtain a summary of actual features, predicted features, and visualization.

### Database

Dataset: https://www.kaggle.com/datasets/jonathanoheix/face-expression-recognition-dataset