Curvelet-based method for orientation estimation of particles from optical images

Jouni Sampo,Jouni Takalo,Samuli Siltanen,Matti Lassas,Jussi Timonen,Arttu Miettinen

doi:10.1117/1.oe.53.3.033109

Abstract

A method based on the curvelet transform is introduced to estimate the orientation distribution from two-dimensional images of small anisotropic particles. Orientation of fibers in paper is considered as a particular application of the method. Theoretical aspects of the suitability of this method are discussed and its efficiency is demonstrated with simulated and real images of fibrous systems. Comparison is made with two traditionally used methods of orientation analysis, and the new curvelet-based method is shown to perform better than these tradi- tional methods. © The Authors. Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution or reproduction of this

Highlights

Orientation analysis of complex patterns is usually done by applying the fast Fourier transform (FFT)[1] or gradient-based methods like the structure tensors (ST).[2,3,4,5] Relative strengths of different orientations are measured, e.g., by investigating the magnitudes of Fourier-transform coefficients usually in polar coordinates
5 Conclusions A new method based on the curvelet transform was introduced for estimating the distribution of orientation in images of elongated features
The known distribution of orientation in the computergenerated network of fibers was accurately produced by this method

Summary

Introduction

Orientation analysis of complex patterns is usually done by applying the fast Fourier transform (FFT)[1] or gradient-based methods like the structure tensors (ST).[2,3,4,5] Relative strengths of different orientations are measured, e.g., by investigating the magnitudes of Fourier-transform coefficients usually in polar coordinates. During the last decade, transforms like the curvelet, contourlet, and shearlet transforms have been developed and have proven to be well suited for various applications.[7,8,9,10] The basis functions of these new transforms are tightly localized in the both space and frequency domains and have in addition a direction angle, i.e., an orientation parameter that makes them promising tools for orientation analysis. We use in this work the orientation of fibers in paper as the basic application and framework. This choice was made because data to be analyzed in this application are common and challenging. If the methods developed work well in this case they will probably work in many other (similar) applications such as, e.g., determination of the orientation of fibers or nanofibrils in reinforced composites.[5,11] in paper-making industry it would be advantageous to have a good orientation analysis method for on-line measurements during the manufacturing process (paper webs move up to 2000 m∕ min)

Methods

Results

Conclusion