ICE2607 Lab3: Scale Invariant Feature Transform (SIFT)

This project simplifies SIFT algorithm with a command-line interface and customizable parameters. It provides the convenience of plotting matching results, gradient visualizations, and similarity heat maps.

File Structure

• src/
  • main.py: Main Python script.
  • plot.py: Contains functions for visualizing SIFT algorithm.
  • sift.py: Contains the implementation of the SIFT class.
• dataset/: Contains default dataset images.
• results/: Contains SIFT matching results of default images.
• figures/: Contains visualizing figures of default images.
• target.jpg: Default target image.
• Lab3_report.pdf
• README.md
• Makefile
• requirements.txt

Example

Matching result

Gradient Visualizations

Gradient Field	Gradient Histogram

Main Orientations	Neighboring Gradients

Similarity Heat Map

Well-Matching Heat Map	Poor-Matching Heat Map

Install

1. Ensure that you have Python installed on your system.

2. Clone or download this project and navigate to the project directory.

3. Check if the dependencies listed in requirements.txt are installed. If not, create a virtual environment (if necessary) and run make install.

Usage

Default SIFT

python -m src.main \
    --image-dir path/to/dataset/images \
    --target-dir path/to/target/image \
    --output-dir path/to/output/folder \
    --output-type png

Note:

• image-dir can be path to the dataset image or the folder containing dataset images.
• Avoid duplicate filenames in image-dir.
• If the parameters are not specified, the program will utilize target.jpg as the target image, five images in the ./dataset directory as dataset images and save all results in PNG format to the ./results directory.
• If you want to use OpenCV's implementation, add --opencv True.

Visualization

Preparation

from src.plot import *
image1 = cv2.imread("path/to/your/image1.jpg")
image2 = cv2.imread("path/to/your/image2.jpg")
print(find_corners(images1))

Use find_corners() to find the coordinates of keypoints.

Gradient Field

plot_local_grad(image1, (x, y), type='figure')

Gradient Histogram (Polar)

plot_local_grad(image1, (x, y), type='hist')

Main orientations

plot_global_grad(image1)

Neighboring Gradients

plot_neighbor_grad(image1, (x, y))

Similarity Heat Map

image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2GRAY)
image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2GRAY)
sift = SIFT()
kp1, dst1 = sift.detectAndCompute(image1, corner_num=100)
kp2, dst2 = sift.detectAndCompute(image2, corner_num=100)
plot_similarity(dst1, dst2)

Note:

• Replace (x, y) with one of the keypoint coordinates obtained during the preparation step.
• Certain plotting parameters, such as the size of arrows, may require manual adjustment based on your input image.