BROWSE
Detail
Conference Paper
Depth Maps Created from Blur Information Using Images with Focus at Near and at Far
Cited 6 time in 
Share 
Share
- Authors
- Suk Hee Cho, Wa J. Tam, Filippo Speranza, Ron Renaud, Nam Ho Hur, Soo In Lee
- Issue Date
- 2006-01
- Citation
- Stereoscopic Displays and Virtual Reality Systems XIII (SPIE 6055), v.6055, pp.1-12
- Publisher
- SPIE
- Language
- English
- Type
- Conference Paper
- Abstract
- A method of producing depth maps for depth-image-based rendering (DIBR) of stereoscopic views is proposed and tested. The method is based on depth-from-defocus techniques, utilizing two original images, one with the camera focused at a near point and the other with it focused at a far point in the captured scene to produce depth maps from blur at edge locations. It is assumed that the level of blur at an edge reflects the distance it is from the focused distance. For each image, estimates of the level of blur edges at local regions are obtained by determining the optimal scale for edge detection based on a luminance gradient. An Edge-Depth map is then obtained by evaluating differences in blur for corresponding regions in the two images. This is followed by an additional process in which regions in the Edge-Depth map that have no depth values are filled to produce a Filled-Depth map. A group of viewers assessed the depth quality of a representative set of stereoscopic images that were produced by DIBR using the two types of depth maps. It was found that the stereoscopic images generated with the Filled-Depth and the Edge-Depth maps produced depth quality ratings that were higher than those produced by their monoscopic, two-dimensional counterparts. Images rendered using the Filled-Depth maps, but not the Edge-Depth maps, produced ratings of depth quality that were equal to those produced with factual, full depth maps. A hypothesis as to how the proposed method might be improved is discussed. © 2006 SPIE-IS&T.
- KSP Keywords
- Depth Map, Representative set, Stereoscopic image, depth quality, depth-image-based rendering(DIBR), edge detection, optimal scale, two-dimensional(2D)