SEGMENTATION OF 2D AND 3D TEXTURES FROM ESTIMATES OF THE LOCAL ORIENTATION

Abstract

We use a method to estimate local orientations in the n-dimensional space from the covariance matrix of the gradient, which can be implemented either in the image space or in the Fourier space. In a second step, two methods allow us to detect sudden changes of orientation in images. The first one uses an index of confidence of the estimated orientation, and the second one the detection of minima of scalar products in a neighbourhood. This is illustrated on 2D Transmission Electrons Microscope images of cellulose cryofracture (to display the organisation of cellulose whiskers and the points of germination), and to 3D images of a TA6V alloy (lamellar microstructure) obtained by microtomography.

Keywords

• 3D image analysis
• covariance matrix
• Fast Fourier Transform
• gradient
• oriented texture
• watershed segmentation