Foreign objects in deboned poultry can be detected by Xray imaging, based on their distinct Xray absorption.The simplest way to segment foreign objects in Xray images is by thresholding, an operation that often fails due to the uneventhickness of the meat fillet. To overcome this problem, we propose an adaptive thresholding method for image segmentation,with the threshold functions resulting from smoothing Xray images. We derive the relations between smoothing window andobject sizes. Because of the adaptiveness to local intensity level, which is proportional to the meat thickness, this techniquecan implement thicknessinvariant image segmentation. Relevant theory and experiment results are reported.