Single View Modeling

Xuewen Chen

In this assignment, I constructed 3-D models for two different images, a real scene and a painting, based on the technique described in paper written by Criminisi et. al. as well as our class notes.

Part 1: house model

First, I scanned a photo from the Journal " Architectural Record" (vol. 86, no. 4, p114, 1998) as shown in figure 1. I chose this house interior since I thought it would be easier to find parallel lines in X, Y, and Z direction. But soon I found it was not a good choice since almost all planar patches I needed were occluded.

Figure 1 original image (house)

I used the method described by Collins to compute vanishing points. Figure 2 showed the lines I used to compute vanishing points.

Figure 2 house with marked lines

Figure 3 is the reconstructed scene, far away from the input image.

Figure 3. a still image of a new view

To construct the 3-D model, I used eleven patches. Figure 4 shows one of the texture (back wall) I used.

Figure 4. back wall from input image

As can be seen, the back wall was occluded by sofa and human heads. I used photoshop on PC to edit the text image: mark the sofa and heads and replace them using textures in input image. Figure 5 shows the texture map after retouching.

Figure 5. back wall after retouching

Following image shows a reconstructed 3-D model .

Figure 6. a reconstructed scene

Figure 7 shows the user interface.

Figure 7. user interface

Figure 8 shows the VRML model with camera.

Figure 8. scene with camera

Corresponding VRML files (with and without camera position) can be found at bottom.

Part 2: painting

In this part, I scanned an italian painting, "The Fano Alterpiece" by Perugino from the book " A chronicle of Italian" (page 123), as shown in figure 9.