Research On Auto Role Replacement Of Video Images

Although it is not long since the film was born,it has been important impact on daily life. With the rapid development of computer technology, computer software can make some frames those can't be worked by the traditional film tenology. Digital image processing technology will also make more perfect effect to film. This paper discusses the role replacement of special effects film which belongs the type of Supplement and Synthesis. It needs to fusion of image processing, pattern recognition and film special effect processing. The audience's digital image is input to computer. The face is located and feature data are extracted from the image. The head pose of image from the camera is replaced by the video image's. Then the face data of video image can be replaced by the audience's.This paper includes some work as follows:Some digital image processing techniques are introduce in chapter II. These techniques are edge detection, contour extraction ,threshold segmentation and contour tracking. There are three methods of edge detection they are Sobel detection, Isotropic Sobel edge detection and LOG edge detection. The processed results are given corresponding these image processing methods. All of the work is mainly prepared to locate the camera image's face. Some face located methods are introduce in chapter III. We locate the face using the largest gray. Then we can locate the positions of eyes, pupils and nostrils, etc. The eyes'position need to record. The face poses include rotation of X, Y and Z axis can be amended and the amendment methods are given out. Currently, the processing mothods of rotation of X axis are very complexed. It is overcomed at a certain extent by placing camera. We placed 9 cameras in front of the audience and these cameras are palced in the same vertical plane. They are belongs in the same arc which the central point is the middle two eyes. There are 15 degrees between rays of every two cameras to central point. The image from a camera is corresponding a level of rotation of X axis. The levels are defined as follows: the image from the camera which it is the same horizon with the audience'eyes when the head pose parallels with vertical direction are defined 5. Then we can define one level every 15 degrees to the former level's vertical dirction. We can divide up rotation and down rotation of Z axis into 4 levels, separately. The largest degree to vertical direction is 60 degrees. These images taken by 9 cameras are belongs 9 levels whchi from 1 to 9 level. To rotation of Y and Z axis, we can adjust the image by the formula from conferences. The vertical direction parallel with the head posture which is a horizontal line in the eyes of the camera to take images defined as the first five, and then the vertical direction and every 15 degree angle definition of a level, the rise and bow to include four-level, the direction of the head with the largest vertical angle in absolute terms is 60 degrees, so nine cameras to take images of a total of 1-9, including a total of nine levels. The rotation and tilt posture, according to the literature, is the attitude adjustment formula.In chapter IV, we introduce the role's face area location methods of video images. The positions of the two eyes and the whole face feature area are given out manualy. Then the distance of two eyes can also be computed. The face featrue model can be obtained and the face poses can be estimated.In chapter V of this paper, we describe the role replacement and give out the results in detail. The role's replacement is done between audience's image and video image. During the replacement, we should adjust the camera image by the face feature information of video image. The the camera image scale is adjusted with the distance of two eyes. By the face model data and the position of left eye's pupil, we replace the pixels of the camera face area with the video image's. Some problems are found during the processing work. Because the degree of the rotation of Y and Z axis of the head poses from video image are estimated, and the rotation of X are divided into 9 levels, the compensation to the poses will produce some errors. If the video image face scale is larger than the camera image's, the role's replacement by the face model will include the camera image background. It also effcts the replacement result. When the eyes see the different direciton, the pupils'position is not same to eyes, it alse affects the results. The diffenrence of skin color will bring the colors in forehead and cheaks. If we use the whole face area to create the face model, the hair on the forehead will affect the result also. During the process, we didn't consider the face expression. In all of the video images, we use almostly the same face expression image to replace the image data.In chapter VI, we conclude the work and point out the shortcomings and future work in this paper. We only make a preliminary discussion to role's replacement. In future work, we will compute the video image's head poses information using some methods. We can avoid the small face area's affect using the background generation. We can also use the whole head to replace the role's image data. The whole camera image head information is used to generate feature model. The more face in origin video image and the background of camera image will be became the background of the the video one. We not only use the position of eyes, but also use other face organs'position duiring the replacement. The problem can be disposed by using the whole head information. The skin color difference can be adjusted into same color also. To the color problem, we can avoid it by the whole head information. We can use face expression analysis and synthesis algorithms to improve the replacement video image fidelity. Some humour results can also be added to the video images. We can use auto face recognition in the video image to the role's replacement. When we extract the feature from the video image, we can use auto face locatation and recognition. Then the role of the video image can be auto found and the feature data can be auto obtained. Because the speed of replacement is raised, the number of the video will be significant increased, the chance to attend into the film for the audience will be increased also. It will stimulate the viewers'greater enthusiasm to films and also promote the digital film technology effectively.In short, this paper is only made a preliminary analysis and study about the role replacement of video images with camera images. We hope it can cause concern and atrract more and more people into this work. After constantly added and updated technology, digital film special effect technology will get new vitality then it will have a future promote development.