Media TechnologyResearch Centre
Shrink to Fit: Image Results
After the Utah Teapot we now have the Glasgow Cup! On the left is an image of an inverted cup. We analysed this for regions of similar texture, segmented the image to continuous boundaries, and then re-rendered it (on the right). The two are virtually identical.

The continuous form is pixel independent, more like a graphical image represented mathematically than a pixel image. We can therefore arbitrarily rotate, scale, shear, re-light, composite and otherwise manipulate the image without the quality problems of pixel images. All on a desktop PC. [Thanks to John Patterson at Glasgow for the cup image.]

The following are samples, produced by Dr Frédéric Labrosse and Dr Maxime Froumentin at Bath, when they were employed here on the first phase of the project.

The remaining examples show the ability of our IRCS software to compose multi-layer images of various kinds. IRCS was used first on its own to depict the virtual rostrum setup. Quasi3D was then used to show that it correctly handles cel intersection. Other, more sophisticated effects such as image warping, zooming and NPR were demonstrated using the image processing utilities.

#### IRCS

The basic version of IRCS was used to render the image that depicts the this virtual rostrum setup (left), with resulting output (right):

This test shows geometric primitives, transformed in perspective and then layered to produce a composite image.

#### Quasi3D

In this case, we test the abilities with pixel-textured rectangles. Four separate stacks are created for producing four separate images (three images with one cel per image and another composite image showing all the cels). Appropriate 4x4 matrices are created for rotating the cels at different angles. The cels are created, applying the transforms and setting their depth in the stack. Having specified each of the photographic textures by name, three rectangles are drawn for assigning them to.

The objects are created from the primitives, adding them to the appropriate cels. The cels, in turn, are added to the appropriate stacks which are saved to file. These are then used for formulating the image information that is used to render three separate images and one extra composite.