Reconstructing 3-D Facial Shape Based on SR Imagine

Although many recent techniques for automatic a 3D facial shape reconstruction from a captured image or the several frames of moving picture have been developed in this regard, the re-construction result of a 3D facial shape in most of these tech-niques significantly depends on resolution of a corresponding image or moving picture. Thus, when a frame resolution of an image or moving picture is low, the reconstruction quality of a 3D facial model can be very poor, resulting in difficulties of us -ing the 3D model for face recognition or an investigation of a suspect.Typically, a reconstructed 3D facial model accurate enough to be used for face recognition or an investigation of a suspect may be obtained only if a human face in a close-up frontal posi-tion is generated with high resolution from a frame of a captured image or moving picture. Thus, the scope of using these tech-niques is very limited. In general, since it is difficult to acquire images of high quality by using cameras in access control de-vices, ATMs, CCTV monitoring devices, and so forth, the sus-pect’s face constitutes a small percentage of the entire frame of an image or moving picture captured in a general state, a facial region usually has a low resolution, and thus, it is difficult to use captured images for an investigation of a suspect. In other words, the traditional studies of reconstructing 3D facial shape is only practical in police investigation under very specific environment condition, for example, the video is high quality and focused on the face.In this paper, we propose a new method to overcome these previous limitations mentioned above. First, we apply the pro-posed SR technique on the sequential video frames which are selected manually including face area in consequence of the accuracy of motion estimation increases in case of moving ob-jects. Second, we reconstruct a 3D facial model with the simple and fast recursive optimization technique based on adaptively selected 8 facial basis models according to facial statistical data. Thus, we can provide not only facial images of various points of view including a frontal view of image which is very useful for face recognition but cannot be provided by a single point of view camera if not captured in that view.