Our long-term customer—for whom we successfully realized a project to create a huge photo catalogue consisting of tens of thousands person profiles— set a new task to create a 3D model for each profile based on the profile photographs.
Usually, 5 to 15 photos are required to create a 3D model of a head. However, this time, we were only given one photograph for each profile, which had been originally captured for a different purpose. That is, each shot was from a different perspective. We were asked to work out whether it would be possible to make a 3D model using only one photo.
First, we calculated the control points on the face, such as the nose, edges of the lips, the ears, the nasolabial folds, and the shape of the face.
Then, we defined the head shapes based on the shading and color depth, which was necessary for the 3D-model creation.
An image was pre-processed to map on the calculated model.
As we had no opportunity to take another photo for the profile, we had to use photos from the existing database of original photos. Because the photograph was of insufficient quality and size, the resulting 3D model could be poor, with some asymmetry or ruptures in the cheek, lip, ear, and nose areas. In that case, an acceptable 3D model could not be obtained without replacing the original photo.
As the customer’s photo catalogue contained tens of thousands images, the workload was huge. Thus, the process needed to be fully automated to avoid unfeasible labor costs. It took three weeks to process all the control points, and as much again to create the 3D models.