Segmentation is usually performed in the spatial domain and is likely hindered by similar intensity, intensity inhomogeneity, and partial volume effect. In this article, a visual-selection method is proposed to carry out segmentation in the intensity space such that the aforementioned
difficulties are alleviated and better results can be produced. The proposed procedure utilizes volume rendering to explore the input data and builds a transfer function, encoding the intensity distribution of the target. Then, by using this transfer function and image processing techniques,
a region of interest (ROI) is constructed in the intensity field. At the following stage, a texture-based region growing computation is conducted to extract the target from the ROI. Experiments show that the proposed method produces high quality results for a phantom which is composed of plates
with similar intensities and textures. It also out-performs a traditional segmentation system in separating organs and tissues from a torso CT-scan data set.