Real-Time Image Mosaicing

By stitching together multiple images taken from different viewpoints or camera orientations, Image Mosaicing technique enables us to experience a photorealistic virtual environment without really having the 3D model of the scene.
I implemented an Image Mosaicing system based on Richard Szeliski's planar perspective motion models. Two Images can be initially stitched together by user interaction, and then automatically stitched in real time. For speed purpose, I exploited hardware-accelerated texture mapping to resample the images and OpenGL Mipmaps to implement hierarchical matching. With this system, I generated the following two 8-image mosaics of University Center Mural:
 

The sixteen images in the first and fifth rows are taken inside University Center. The mosaics following them are generated pair wise. The final mosaics in the fourth and eighth rows are of size 827x311 and 1008x488 respectively. It takes about 1 second to generate each of them.
Reference:
R. Szeliski and H.-Y. Shum, Creating full view panoramic image mosaics and Environment Maps, Computer
       Graphics (SIGGRAPH'97), pages 251-258, August 1997.

2.5D Image Warping

The geometry of the real-world is exceedingly complex. Light's interaction with the geometry is hard to simulate. However, images taken from the real-world capture the result of this simulation. Given an image with the camera calibration, new images from novel viewpoints are able to be generated  by image warping.
I implemented an Image Warping system based on Leonard McMillan's warping equation with visibility ordering algorithm. I used Vivid Scanner to produce a 3D model of my face, ( too timid to open my eyes ) and used Paul's codes to convert it into ( R,G,B,Delta ) format with camera information given. I generated 8 images from new viewpoints, shown as follows:
 

The the image in the middle is taken by Vivid Scanner. The images around it are synthesized by image warping.
Reference:
Leonard McMillan's Ph.D. Thesis.

Image Morphing

Subdivision Surface

Recursive Ray Tracing

The following picture is rendered by my simple recursive ray tracer that includes shadow casting, specular reflection, and refractive transparency.

 

Radiosity Method

return to Li Zhang's homepage