Abstract: Skin region detection plays an important role in a variety of applications such as face detection and adult image filtering. To improve the accuracy and speed of skin detection, an adaptive skin detection algorithm working on compressed domain of JPEG images is proposed. Two main contributions of the skin detector are: ① It can adaptively control the threshold according to image content; ② It doesn't require full decompression of JPEG images. It directly derives color and texture features of each image block from DCT coefficients and jointly uses both features to classify each image block as human skin region or not; Comparisons with other existing skin detection techniques demonstrate that the algorithm proposed can compute very fast and achieve good accuracy.