In this paper, we study low-level image segmentation in the normalized cut framework proposed by Shi and Malik (1997). The goal is to partition the image from a big picture point of view. Perceptually signicant groups are detected rst while small variations and details are treated later. Dierent image features-intensity, color, texture, con-tour continuity, motion and stereo disparity are treated in one uniform framework. We suggest directions for intermediate-level grouping on the output of this low-level segmentation.