Seam carving is a technique for resizing an image while intelligently preserving "important" content such as people or structures.
Shai Avidan and Ariel Shamir published a SIGGRAPH 2007 paper describing the algorithm.
When an image needs to contract, the algorithm finds a connected "seam" in the image running edge to edge along a "path of least importance."
Importance is determined by an energy function that assigns a cost to each pixel.
The code illustrates exposes phase of the algorithm as a separate image:
- Conversion to grayscale.
- Computation of the gradient magnitude.
- Finding the minimal seam.
- Carving out the seam.
It makes extensive use of the HTML5 canvas element.
The code is not geared toward efficiency, but rather, human readability.
My goal was to make the algorithm itself as clear as possible.
The tower image is from Wikipedia.
