Professors Shai Avidan and Ariel Shamir recently demonstrated a new image resizing technique titled "Content Aware Image Resizing".
Today a simple html tag can declare a text entry to be a percentage of the window's width or height, allowing the text to be elegantly wrapped on window resize. Fonts are also easily adjustable in size via the control key and the scroll of a mouse button.
While this has made browsing a much more pleasant experience, images unfortunately have been unable to scale successfully in modern browsers, due to the complexity of shrinking or enlarging an image, without losing data or creating artifacts.
Some image resizing techniques do exist, such as bilinear or bicubic filtering, but they tend to sharpen or smooth an image, and are computationally intensive.
Professor Shai Avidan, funded by Mitsubishi Electric Research Lab, and Ariel Shamir, working for and funded by the The Interdisciplinary Center & MERL, have created a promising new technique for image resizing. They demonstrated their technique at San Francisco’s SIGGRAPH 2007 conference, which they have demonstrated in a video on YouTube.
The video demonstrates them employing their technique to easily resize images. The technique involves using gradients to calculate a pixel path of least importance. This pixel path selects one pixel from every row of pixels to remove. By using gradients, the impact on the image's content is minimized, effectively.
Further, they demonstrate the shortcomings of their application, and how they have developed a resizing program utility to overcome them. One such limitation is faces, which compress poorly and are prone to artifacting. Their utility compensates for this by providing a protection brush that allows you to paint with a transparency brush over portions of the image you want to protect. These areas are automatically rejected by the pixel path routes, so are not removed or resized.
Further, a side application of the technology demonstrated was the ability to use a different color transparency brush to designate certain areas to always remove. Whether the user is a jilted lover, looking to remove their significant other from certain pictures, or the US or Chinese governments looking to remove images of their weapons, discretely, from the public eye, this technology may have many uses.
While solutions like this have existed before, they typically look much more noticeable than the subtle, almost unnoticeable, removal that the "Content Aware Image Resizing" technique accomplishes.
For the paper about this technique titled "Seam-Carving for Content-Aware Image Resizing", or more of Dr. Shamir's diverse science and technology related research, please view the professor's website.
While the technique is likely computationally intensive, it may provide something to put tommorrow's quad core chip powerhouses to work, in creating more dynamic media content.
Hopefully, this technique will be coming soon to a browser or imaging suite near you.
|