Overview
This project compares three approaches to classifying facial images into age groups. It examines traditional computer vision pipelines alongside a convolutional neural network, using the same dataset to compare their behaviour.
The task is age-group classification rather than predicting an exact age.
Approaches compared
SVM with HOG features
A Support Vector Machine uses Histogram of Oriented Gradients features extracted from the images. This approach separates feature extraction from classification.
Bag of Visual Words with SIFT
A Bag of Visual Words approach represents images using SIFT-based local features, providing another way to turn visual information into inputs for classification.
Convolutional neural network
A CNN is trained end to end, learning image features as part of the classification model rather than relying on a separately designed feature representation.
Evaluation
Each approach was trained and evaluated using the same dataset. The comparison considered classification accuracy and computational performance, with attention to the trade-offs between feature-based pipelines and deep learning.
The aim was to understand the different approaches in the context of the same task, rather than assume that one method would be the best fit in every setting.
What the project demonstrates
The study brings together image processing, feature extraction, model training and comparative evaluation. It provides practical experience with both traditional machine learning and neural network methods for computer vision.
Status
Completed.