Research Of Digital Image Processing Algorithms In The Old Movies' Restoration

Movies or videos are common methods for recording large scale social activities. Our country's old movies are videos recording century history, and have very important historical education meaning. Most of old movies are stored in the form of films, because of careless and many times of play, film damages are generated. So, the restoration of old movies is urgent.When old movies are restored using restoration software, the region of human face may be blurred after filtering and de-noising. But when audiences watch movies, human face is the position of most concern. So, human face need to be treated carefully in old movies restoration. The premise of face special handling is face detection and location. This thesis researches this problem using image processing algorithms.The blurring phenomena are obvious in bigger faces, this thesis detects bigger human faces in old movies. For improving the speed of face detection, face detection is combined with face tracking. Face tracking may causes face losing when shot change takes place, so shot segmentation is needed. First, shot segmentation is done in old movies, and then face detection in the first frame picture of a shot, at last face tracking in the next continual picture on the basis of face detection in the same shot.The methods of shot change include abrupt change and gradual change, this thesis researches shot abrupt change detection and fade-in and fade-out of gradual change detection. Face detection uses ellipse fit algorithm according to the feature of contour, firstly face detecting is made by means of least square ellipse fit, but through experiment we find that this algorithm only has good results in ideal pictures, so improvement is made by least square ellipse fitting algorithm. And it can gain ideal results, but the speed of improved method is slow. Human face tracking in the next continual picture utilizes Mean Shift algorithm, and binding with the characteristic of gray-level and take gray-level direction code as the character in the process of constructing target mode, through the experiments achieved ideal results.