Segmentation Pre-processing

With Asymmetrical Blurring

By Timothy Bahls

Abstract:

I will describe a robust segmenting pre-processor I have implemented by improving methods worked on by D. Scharstein and R. Szeliski. The image will be iteratively blurred, increasing the differences between segments while decreasing the differences inside a single segment. The resulting image can then be segmented by rather simple segmenter because the results will be nearly completely segmented. This segmenter improves the speed and effectiveness of the blur by implementing a larger window size, and by segmenting a scaled version of the image.

The segmenter has been upgraded considerably since the presentation. It now uses a scaled pyramid of images to segment. The segmenting takes a little bit longer, but it now spreads colors across segments much more effectively and is much better and keeping together noisy segments.

The paper can be downloaded here.

The Java code can be found here.

Typical results are given below.

Results

A special thanks to Magda, my amazing, talented, and beautiful fiancée for putting up with my using pictures of her for class.