Interesing Papers in ICCV09

 1. Kernel Methods for Weakly Supervised Mean Shift Clustering

Combine the prior information to improve Mean Shift Clustering, which is supervised method. A simple example of 1 dimension data clustering is in figure1.

 

2. Filter Flow

This paper considered the filter flow problem which is computea space-variant linear filter that transforms one image into another, including Affine transformation, depth from defocus, etc. The auther made the conclusion that "any combination of the linear constraints and objectives can be solved to obtain the globally optimal solution using Linear Programming(LP) based on some constraints. It is an inverse problem to find the transformation between two similary images.