Content Aware Resize

The Seam Carving for Content-Aware Resizing was proposed by Shai Avidan and Ariel Shamir at SIGGRAPH 2007. The image is scaled by adding or removing seams that have a low amount of detail. The result is a resize that can be tuned to avoid distorting objects of interest like people.

Overview

Seam Carving is performed by identifying scan lines in the image that require a minimal amount of energy to traverse. Energy in this plugin is user defined, however, the most logical choices are Sobel edge detection or Laplacian of Gaussian images. To improve performance the horizontal and vertical seam carving are alternated based on a resampling frequency.

Inputs

Results

The original input image.

The width of the image is increased from 787 pixels to 987 pixels. The man is distorted since the pixels in his body contain little detail.

A manual energy mask can be defined using the Sobel filter and paint tool. Objects that should not be resized are painted with solid white.

The image resized using the manual energy mask. The image is extended but the person is not distorted.