There are huge number of audio visual content on the Web site like YouTube or NicoNicoDoga and the number has been expanding drastically day-by-day. To find and get the one’s most favorite music or video content quickly and efficiently is one of the most important topic in data retrieval research area. Also as CGM (Consumer Generated Media) is focused recently, an amateur or novice user can have a chance to produce and upload his own content to Web site and these are becoming a new cultural phenomenon. So the needs to a content authoring tool for amateur users expected to be increasingly expanding. This project is strongly supporting this need by producing innovative technologies and authoring tools considering the drastic change of a style of content production and content viewing environment.
This is one of the core projects of Morishima-lab. in which Future Cast System is drastically updated and advanced to be a very affordable technology. A face geometry estimation method with light calculation cost and high accuracy only from a single picture using face prior from a collected face database is proposed at first. Then from video captured face image, a much more accurate 3D face geometry can be reconstructed by utilizing a Structure-from-Motion method to calculate 3D coordinates of sparse feature points as well as a former 2D feature points and face priors of frontal face image. And now, new trial is performed to achieve a higher depth accuracy by utilizing a normal information calculated by Shape-from-shading or by utilizing only a diagonal face silhouette curve information.