About Scale Space
The Scale-Invariant Feature Transform (SIFT) algorithm detects local features in images. To ensure features are found regardless of their size, the image is progressively blurred using a Gaussian filter, creating a "Scale Space".
By taking the Difference of Gaussians (DoG) between adjacent blurs, we can identify "characteristic scales" — spatial and scale coordinates where the DoG response peaks. This tool lets you visualize this process interactively.
Controls
No file chosen
3
1.6
Waiting for image...
Source Image (Click to analyze point)
Characteristic Scale Responses
Click on a point in the source image to plot its DoG response across different scales. The peak of this plot defines the characteristic scale. Blue rings will be drawn over your click to visualize the physical area (where radius corresponds to the blur amount σ) of each detected peak.
Load an image to view the Gaussian Pyramid
Load an image to view the DoG Pyramid