Automatic Data-Driven Parameterization for Phase-Based Bone Localization in US Using Log-Gabor Filters

Intensity-invariant local phase-based feature extraction techniques have been previously proposed for both soft tissue and bone surface localization in ultrasound. A key challenge with such techniques is optimizing the selection of appropriate filter parameters whose values are typically chosen empirically and kept fixed for a given image. In this paper we present a novel method for contextual parameter selection that is adaptive to image content. Our technique automatically selects the scale, bandwidth and orientation parameters of Log-Gabor filters for optimizing the local phase symmetry in ultrasound images. The proposed approach incorporates principle curvature computed from the Hessian matrix and directional filter banks in a phase scale-space framework. Evaluations performed on in vivo and in vitro data demonstrate the improvement in accuracy of bone surface localization compared to empirically set parameterization results.