Project Mata Kuliah Artificial Intelligence - Face Expression Recognition

Short Explanation

The "Face Expression Recognition" project aims to recognize human facial expressions using the Convolutional Neural Network (CNN) method. The CNN algorithm is applied to analyze visual data such as facial images in grayscale format, which are then classified into seven basic expression categories: happy, sad, angry, surprised, afraid, disgusted, and neutral. This model was trained using the FER2013 dataset and managed to achieve an accuracy of 91.67% after training for 500 epochs.

Project Goals

This "Face Expression Recognition" project is the final project of the Artificial Intelligence course where in this project there are achievements that must be achieved including:

1. Developing an artificial intelligence-based facial expression recognition system. This system is expected to be able to identify emotions radiating from facial expressions automatically and accurately.
2. Experiment with machine learning algorithms to improve facial expression recognition accuracy. In this project, the CNN algorithm is tested to understand the extent to which this model is able to recognize complex patterns in facial images. This effort also includes optimizing model parameters, adding training data, and using data augmentation methods.

Tech Stack Used

1. Framework: Python uses libraries such as TensorFlow/Keras for CNN implementation.
2. Dataset: The dataset used is FER2013 (Facial Expression Recognition 2013), which contains 35,887 grayscale images of faces with dimensions of 48x48 pixels. These images come with labels covering seven basic expression categories.
3. Tools: 

• NumPy and Pandas for data manipulation.
• Matplotlib for visualization.
• Haar Cascade for face detection from camera.

Results

1. Happy
2. Sad
3. Angry
4. Neutral
5. Surprised
6. Afraid
7. Disgusting

The Problems and How I Deal With It

1. The problem of differences in lighting which affects the level of accuracy. 
   Lighting variations can affect model accuracy. To overcome this, data normalization is carried out to ensure that the lighting in the image is more uniform so that patterns in facial images can be recognized better.

2. Similar complexity of expressions.
   Some expressions, such as “scared” and “surprised,” have similar characteristics that are difficult for the model to differentiate. The solution implemented is to carry out data augmentation such as rotation, zooming, flipping, and contrast changes to increase the generalization ability of the model to new data.

3. Quite limited dataset
   The FER2013 dataset, although quite large, does not cover the full range of face variations globally. To enrich the dataset, I used data augmentation techniques and added data from other relevant sources to create a better representation of facial expressions.

Lessons Learned

This project provides deep insight into how artificial intelligence-based systems can be used to recognize facial expressions. The development process shows the importance of:

1. Data pre-processing to address lighting issues and improve data quality.
2. Experiment training parameters to get the optimal combination, such as setting the number of epochs, learning rate, and batch size.
3. Increased diversity of training data through augmentation to improve model performance against real-world data.

By overcoming existing challenges, this project succeeded in building a facial expression recognition model that can be applied to various applications such as human-computer interaction, emotion analysis, and psychological monitoring.

The above is the detailed content of Project Mata Kuliah Artificial Intelligence - Face Expression Recognition. For more information, please follow other related articles on the PHP Chinese website!