An apparatus and method for detecting a head image of an input image. In the head image detection apparatus, a training DB stores therein a positive head image and a negative head image into which a plurality of images are classified. A sub-window processor extracts a feature population while sliding a sub-window of a predetermined size with respect to the input image. A head candidate region classifier classifies head candidate regions based on the extracted feature population from the sub-window processor by referring to samples stored in the training DB. A head candidate region determiner determines a head candidate region as a head region by checking continuity of contours of the classified head candidate regions.