The invention relates to a clothes advertisement putting method and
system based on video leading role identification. The method comprises the following steps: A, face identification, i.e., obtaining such information as face pictures, coordinate positions and the like in video pictures; B, face tracking, i.e., gathering face pictures of the same person in continuous video frames so as to realize preliminary clustering of faces; C,
face verification, i.e., calculating a distance between every two face pictures; D, face clustering, i.e., gathering the face pictures of the same person, and finding major roles in a video; E,
human body detection, i.e., obtaining correlation information of a
human body in the video for each frame of the video pictures by use of an area
convolution feature; F, clothes marking, i.e., classifying and marking clothes in
human body pictures; G, background removal and human body segmentation and clustering, i.e.,
cutting human bodies from the pictures, removing backgrounds unnecessary to the human bodies, and segmenting the human body clothes into multiple areas; and H, clothes picture retrieval, i.e., retrieving advertisement objects searching for matched clothes in a
database by use of the clothes.