Short explanation

Project "Face Expression Recognition" 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 is then classified into seven categories of basic expressions: happy, sad, angry, shocked, fear, disgust, and neutral. This model is trained using Dataset Fer2013 and successfully reached an accuracy of 91.67% after training for 500 EPOCH.

Project Goals

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

1. Developing a system of introducing facial expression based on artificial intelligence. This system is expected to be able to identify emotions that radiate from facial expressions automatically and accurately.
2. experimenting with machine learning algorithms to increase the accuracy of facial expressions. In this project, the CNN algorithm is tested to understand the extent to which this model is able to recognize complex patterns in face drawings. This effort also includes optimizing the model parameters, additional training data, and the use of data augmentation methods.

Tech Stack Used

1. Framework: Python uses library like Tensorflow/hard to implement CNN.
2. Dataset: The dataset used is Fer2013 (Facial Expression Recognition 2013), which contains 35,887 face grayscale images with dimensions of 48x48 pixels. These images are equipped with a label that includes seven basic expression categories.
3. Tools:

• Numpy and Pandas for data manipulation.
• Matplotlib for visualization.
• Haar Cascade for the face detection of the camera.

Results

1. Like
2. Sad
3. Angry
4. Neutral
5. surprised
6. Afraid
7. Disgusting

The problem and how I deal with it

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

2. complexity of similar expressions.
   Some expressions, such as "fear" and "surprised," have similar characteristics that are difficult to distinguish by models. The solution applied is to carry out data augmentation such as rotation, zoom, flipping, and contrast changes to improve the ability of models generalization of new data.

3. Dataset which is quite limited
   Dataset Fer2013, although quite large, does not cover a variety of facial variations globally. To enrich the dataset, I use the data augmentation technique and add data from other relevant sources to create a better representation of facial expressions.

Lessons Learned

This project provides in -depth insights on how artificial intelligence -based systems can be used to recognize facial expressions. The development process shows its importance:

1. Data PRA-PER-PERMESSESIA to handle lighting problems and improve data quality.
2. Experiment of training parameters to get optimal combinations, such as regulating the number of EPOCH, Learning Rate, and Batch Size.
3. Increasing the diversity of training data through augmentation to improve the performance of the model of real world data.

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