This paper presents a new nude image detection method. The foreground-background co-sampling mechanism is proposed to online sample the target image's chromatic distribution. The sampled data is then fed into a multiplayer perceptrons neural network to decide if the target image contains skin color object and which color category it belongs to. The texture feature, coarseness, is also utilized to acquire accurate skin segmentation. Several low-level whereas reliable geometrical constraints are further employed to inspect the skin regions. Finally, the mug shot exclusion procedure is applied to promote the system performance. The experimental results show our method is satisfactory for nude image detection.