Go to  Advanced Search

HDR image construction from multi-exposed stereo LDR images

Show full item record

Files in this item

Files Size Format Description   View
ubc_2012_spring_sun_ning.pdf 2.585Mb Adobe Portable Document Format   View/Open
Title: HDR image construction from multi-exposed stereo LDR images
Author: Sun, Ning
Degree Master of Applied Science - MASc
Program Electrical and Computer Engineering
Copyright Date: 2012
Publicly Available in cIRcle 2012-01-18
Abstract: The vast majority of cameras in the market nowadays can only capture a limited dynamic range of a scene. To generate high dynamic range (HDR) images, most existing methods use multiple images obtained from a single low dynamic range (LDR) camera at consecutive instances. These methods can obtain good quality HDR images for still or slow motion scenes but not for scenes with fast motion. In this thesis, we propose the use of two LDR cameras which have different exposures. To generate an HDR image, the two differently exposed LDR images of the same scene are used. The two LDR images should be captured at the same instance, so as to deal with scenes with fast motion. The most challenging step in this approach is to obtain accurate estimates of the disparity maps of the scenes. This will allow us to correctly align the pixels from the two differently exposed pictures when forming the HDR images. Very few state-of-the-art stereo matching algorithms can deal with the problem of obtaining accurate estimates of the disparity map from two differently exposed images. This is because the input LDR images that are used to construct HDR images have large radiometric changes. In addition, the two input LDR images usually have saturations in different areas. To obtain accurate disparity maps, we present a novel algorithm that obtains an initial estimate of the disparity map. Then a refinement step is used to minimize the edge effect and interpolates the values in the saturated regions. Compared to other state-of-the-art methods, our algorithm has a simpler set up with only two standard commercial LDR cameras. The offline processing of the LDR images has a simpler cost function, especially the cost function we use in the refinement step of the disparity map. This reduces the computational complexity and thus the processing time of the LDR images to form the HDR image. Moreover, the disparity map computed by our algorithm can tolerate greater radiometric changes and saturations. Therefore, the HDR images constructed by our algorithm are smoother and have fewer defects than those constructed by other methods.
URI: http://hdl.handle.net/2429/40130
Scholarly Level: Graduate

This item appears in the following Collection(s)

Show full item record

All items in cIRcle are protected by copyright, with all rights reserved.

UBC Library
1961 East Mall
Vancouver, B.C.
Canada V6T 1Z1
Tel: 604-822-6375
Fax: 604-822-3893