Current image-based rendering techniques all share several advantages over geometry-based techniques, making them useful for a wide range of situations. Image-based techniques are independent of the complexity of the scene to be rendered, making real-time realistic interactions with complex scenes possible. In addition, these methods can make use of rendered images, or images of actual physical environments. As a result, complex, realistic real-life scenes can be viewed without difficult or impossible geometric modeling, and independent of scene complexity.

However, current image-based techniques rely on assorted types of sampling and interpolation of source images. Therefore, where the input images are too sparse or when they insufficiently sample the environment, errors may occur. The image-based techniques that use a small number of input images may fail to accurately capture high frequency components and changes in the scene, such as specular highlights.

We are comparing image-based renderings of scenes (both synthetic and real world environments) with a physically accurate image or rendering. These comparisons can be made for both physically-based geometric renderings and measurements of the actual physical environment, if the environment exists and can be accurately measured. We hope to identify the most important sources of error in image-based rendering, and provide algorithms to alleviate the problems they can cause.