1. PCA-SIFT

This section involves:

• Computing PCA-SIFT feature descriptors.
• Modifying images using transformations like scaling, rotation, and Gaussian blur.
• Analyzing keypoints qualitatively and quantitatively.

The process includes:

• Image preprocessing
• PCA-SIFT descriptor calculation
• Matching keypoints between original and modified images
• Result visualization

2. Image Classification

In this section, a custom CNN model is built and trained on the CIFAR-10 dataset. Key tasks include:

• Data preprocessing
• Building a CNN architecture with Conv, Sigmoid/ReLU activation functions, pooling layers, and fully connected layers.
• Evaluating performance using accuracy on test data.
• Comparing results between models using Sigmoid vs. ReLU activation functions.

Images

Images are included throughout the report to demonstrate the process and results:

• PCA-SIFT keypoint detection results
• Image classification accuracy and loss plots