The image to the right (click on it for more related images) is an example of the research work I’m doing this summer. It involves image processing using “wedgelets” — squares that have a line dividing it into two regions, each with a different intensity value. By combining a variety of different line orientations within each square, as well as different scales of squares (not shown here), images can be represented fairly efficiently with relatively little information.

The advantage that wedgelets offer over conventional image representation and decomposition techniques used for filtering/compressing/simplifying images (i.e. 2-D Fourier/Wavelet Transform) is that they are very well-suited to representing image edges, which is typically how the most important information in an image is stored. Another advantage of wedgelets is that geometric information is inherent in a wedgelet representation. This means that once you represent an image using wedgelets, you also know where the edges are in the image, potentially saving you from performing edge detection at a later time.

My next goal is to implement the full wedgelet transform. In addition to having more wedge orientations, this involves having squares of various scales in the image so that more wedges can be used to represent a detailed area of an image, while fewer wedgelets can be used to represent large, slowly-changing areas of an image.

This post was mainly for the engineers in the crowd. However, as usual, a picture is worth a thousand words, so if you don’t understand or need a visual explanation click on the picture above to see the wedgelets used to approximate an image of a star.

This entry was posted on Friday, June 30th, 2006 at 11:53 am and is filed under Geekery, Goings On, School. You can follow any responses to this entry through the RSS 2.0 feed. Both comments and pings are currently closed.