Lior Shapira

I recently returned from ICCV 2009, where I presented a paper by Ariel Shamir, Shai Avidan and myself. The paper is titled ‘Mode-detection via Median-shift’ and deals with very fast clustering of large high-dimensional data sets. Our paper was part of the poster sessions, and it was the first time I’ve presented a poster at a vision conference. It was a very intense experience, over three hours of explaining and demonstrating our work to small groups of people. In the end I was completely exhausted but happy with the level of interest shown. In the project page you can download the paper and the poster, and hopefully soon some version of our software.

I’ve attended most of the sessions and overall I was impressed with the quality of papers presented. My two favorite papers were Super-resolution from a Single Image by Glasner, Bagon and Irani, and Building Rome in a Day by Agarwal et al.

A paper I wrote with Arik and Danny has been accepted to EG ’09. The paper is title ‘Image Appearance Exploration by Model Based Navigation’ and you can see the project page here.

The paper can be broadly categorized as an image processing paper, it deals with changing the appearance of an image by changing its colors. We tackle two major issues. The first one is how to model color changes in an image, and synthesize a new one with no (or little) artifacts. The second is how do we empower a user to make these changes? In most cases the user doesn’t have a well established desired outcome, and even if he does, getting there is a very tiresome trial and error process.

We model color changes using a Gaussian Mixture Model on the image’s colors, and present an innovative gallery based user interface which lets users explore through variations of an image, until they reach a desired result.