Summary

Image-domain approaches emphasize the role of photographs or still-video in order to provide realism to computer generated images. This methodology uses 2D photographic images instead of 3D geometrical models. Realistic virtual views as they are seen from a virtual camera can be generated by an image-based rendering algorithm without any tedious reconstruction of 3D models. Views generated using this method have an advantage compared to those created using the geometrical model-based method, generating the same image quality is much easier. In addition, views generation is relatively easy following "preparation" of the 2D images.

In general image-domain approaches need less computation resource than 3D model-based approaches and the produced image quality is as good as conventional 2D media. However, interaction with the world is limited and they need larger amount of data space, because they have to handle redundant data. Huge amount of data space represents a trade-off for making application involving networks, since a high bandwidth is required to share an image-based virtual world. Also, image-based approaches limit supported virtual views to a "narrow range" and scene is constrained to convex and not occluded objects.

Model-based approaches, on the other hand, are very generic, capable of generating any world and object by using a geometrical model from the beginning. Users encounter no limitations in interacting with the world. Model-based approaches make larger the range of the possible virtual views, and faster the rendering process since they can exploit hardware rendering provided on nowadays graphics-workstations.

Depending on the details and fidelity of the recovered model, model-based methods can yield realistic images. In particular, a realistic synthesis depends on: accuracy of the geometric model of the objects, textures, object surface properties, illumination simulations, rendering algorithms, etc. Pre-acquired information, heuristics and additional effects can also be integrated. A summary of main advantages and disadvantages of the two classes of approaches is presented in figure 3.

Figure: The table summarizes advantages and disadvantages of image- and model- based rendering techniques.

As it comes out from the state of the art, a realistic image synthesis of virtual environments is a large field of applications not yet generally solved. The success of some approaches mainly depends on the application and application constraints. However, from the many different techniques proposed, it is possible to gain a general idea on what approach and on what technique may better fit the individual application dependencies. The parameters playing an important role are:

*

real-time performance;

*

static/dynamic environment;

*

convexity/concavity of the scene objects;

*

object visibility and mutual occlusions;

*

range of perspective transformations required;

*

level of fidelity required.

Several years ago the image-based approaches were unthinkable to propose since memory devices and high speed data links costed so much. Then amazing advances in semiconductor technologies, including reduction of memory device cost, made possible to explore such methodology.

A remarkable progress has also been reached by computer graphics technology, and by 3D geometric modeling. We currently have many sophisticated 3D graphics tools such as 3D modeling systems, 3D scanning systems, that combined with hours of labor let us generate sophisticated graphics images as seen in movies.