It is a fundamental problem to obtain texture information of 3D models in computer graphics. Texture mapping was often used to complete this task. However, texture mapping needed a suitable texture image. Therefore, an interactive colorization algorithm is proposed in this paper to obtain texture information more easily. For a mesh model, salient features are first extracted and classified to reduce user’s interaction on repetitive features, and then user can scribble some curves with desired colors in various regions as seed curves. Secondly, position, normal, and curvature information are combined to measure vertices similarity to avoid color bleeding between neighboring regions, and random walk algorithm is adopted to compute the probability from each vertex to each seed curve. Finally, weighted by these probabilities, the color of each vertex can be expressed as the weighted average of seed curves’ color. Furthermore, the above algorithm is applied to colorize point clouds. Experiment results show that our algorithm can distinguish different regions accurately, and colorize 3D models robustly.