Toward Texture-Based 3D Level Set Image Segmentation

Abstract

This paper presents a three-dimensional level set-based image segmentation method. Instead of the typical image features, like intensity or edge information, the method uses texture feature analysis in order to be more applicable to image sets withs distinctive patterns. The current implementation makes use of a set of Grey Level Co-occurrence Matrix texture features that are generated and selected according to the characteristics of the initial region. The region is then deformed using the level set-based algorithm to cover the desired image area. The generation of the texture features and the level set surface deformation scheme are performed with graphics card hardware acceleration. The preliminary experiments, performed on synthetic data sets, show promising segmentation results.

GPU-accelerated image segmentation based on level sets and multiple texture features

“…where D(p) is the image data term that drives the deformation, C(p) = div(∇φ(p)/|∇φ(p)|) is the curvature and α ∈ [0, 1] is a user-defined balancing parameter. Instead of the original intensity-based image term, we apply a texture-based term D tex (p) [41] that takes into consideration the previously selected texture feature set. First, for a given image point p, a set of features S p is defined as:…”

Multi-Resolution Texture-Based 3D Level Set Segmentation

“…where img(p) is the image intensity in p, while T and are the intensity target value and tolerance: the surface is prompted to expand if img(p) is between T − and T + , and to shrink if it is out of this range. In this work, we employ a multi-feature image term I tex (p) [36] that takes into consideration the features (denoted as a set M ) generated for a given volume resolution. For each surface point p, a subset of features S p is defined as:…”