<<HTML(<meta http-equiv="refresh" content="0;URL=../../CMV/Downloads/MultiScaleAutoconvolution">)>>

Multi-Scale Autoconvolution

Introduction

In many computer vision applications, there is a need to capture the characteristics of the objects depicted so that they are invariant under different geometric transformations. The Multi-Scale Autoconvolution (MSA) method offers a novel way of approaching this problem. The method provides affine invariant features with only moderate computational complexity and does not require any other segmentation steps than background elimination. It is also possible to implement a totally new kind on convexity measure based on MSA. Here we will offer an introduction to Multi-Scale Autoconvolution (MSApage.pdf) and give a sample Matlab programs for computing the MSA values ( msa.m ) and convexity measures based on MSA ( C_12.m ) and ( C_a.m ).

We have also developed another affine invariant technique that is very fast, but lacks some properties of MSA. However it very suitable for application that require high computational performance. We provide example algorithms for computing this invariant. Using first algorithm you can achive the invariant for any alpha beta pair. If you need invariants to be computed using several alpha beta pairs, it is more efficient to implement such algorithm that perform only one interpolation, e.g. the second- or third sample algorithms.

Publications







BiBTeX references (ref.txt)

The Academy of Finland is gratefully acknowledged for providing the funding for the research of the Multi-Scale Autoconvolution (project no. 102083). We would also like to thank the Rolf Nevanlinna Institute for the insightful co-operation in the development of MSA.

MVG/Downloads/MultiScaleAutoconvolution (last edited 2011-11-22 12:20:17 by WebMaster)