Park proposes a fast and stable image-based modeling method that generates three-dimensional (3D) models with high-quality face textures in a semi-automatic way. This means that the algorithm needs a user’s help. The modeler guides untrained users in quickly obtaining 3D model data via several steps of simple user interface operations using predefined 3D primitives. The proposed method contains an iterative nonlinear error-minimization technique in the model estimation step, with an error function based on finite line segments instead of infinite lines.

There is an interesting field of work arising in the use of multiple camera views for scene reconstruction. The features used are usually straight lines, points, curved lines, and so on. By using the algorithm proposed by the author, a user first determines what should be modeled, and provides image corner points for the target model. The interactive modeling system accepts manual input of the corner points, and reconstructs 3D models based on the input measurement data.

A key aspect is the 3D recovery engine. A nonlinear optimization method is used to try to minimize the global error (sum of areas of triangles defined by original and estimated edges). The process is mainly focused on model accuracy, and consists of several steps of numerical computation. Inside the model creation block, and taking efficiency into account, the approach chosen involves the user first manually assigning the line segments that are the edges of the target model on the given images. The system then estimates 3D lines, using an error-minimization technique. Thus, the algorithm uses predefined 3D primitives, like plate, box, and pyramid.

All of the tests described are quite box oriented. Curved geometry could require other 3D primitives, but the system seems easy to extend. Another problem could be the growing number of parameters required in complex geometries. Some studies of scalability could also be interesting.