Image processing apparatus and image processing method

摘要

An image processing apparatus and method may accurately separate only humans among moving objects, and also accurately separate even humans who have no motion via human segmentation using a depth data and face detection technology. The apparatus includes a face detecting unit to detect a human face in an input color image, a background model producing/updating unit to produce a background model using a depth data of an input first frame and face detection results, a candidate region extracting unit to produce a candidate region as a human body region by comparing the background model with a depth data of an input second or subsequent frame, and to extract a final candidate region by removing a region containing a moving object other than a human from the candidate region, and a human body region extracting unit to extract the human body region from the candidate region.

主权项

1. An image processing apparatus comprising:
a face detecting unit to detect a human face in an input image; a background model producing/updating unit to produce a background model using a depth data of an input first frame and detection results of the face detecting unit; a candidate region extracting unit to produce a candidate region that will serve as a human body region by comparing the background model with a depth data of an input second or subsequent frame, and to extract a final candidate region by removing a region, which is judged using the detection results of the face detecting unit as containing a moving object other than a human, from the candidate region, wherein the candidate region extracting unit is configured to set a closer region than the background model as the candidate region; and a human body region extracting unit to extract the human body region from the candidate region that is extracted from the depth data of a current frame using a human body region extracted from the depth data of a previous frame, the human body region extracting unit including,
a first calculator to search the human body region from the candidate region extracted from the depth data of the current frame via implementation of tracking using hard constraint conditions,a second calculator to expand the human body region, searched by the first calculator, via implementation of tracking using soft constraint conditions to complete the human body region, anda third calculator to extract a new human body region via inspection of the candidate region that remains rather than being extracted as the human body region during tracking using the hard constraint conditions used by the first calculator and tracking using the soft constraint conditions used by the second calculator, the new body region being different than the body region extracted from the depth data of the previous frame.