A Framework for Multiresolution Stereoscopic Image Processing

U. Ahlvers, U. Zölzer, and M. Hampe (Germany)


Image pyramids, octave filter banks, disparity estimation, sub-pixel accuracy, stereo vision.


The central task of stereo vision is the estimation of 3D information from stereoscopic image pairs. A drawback of most existing estimation algorithms is the limited value range for the disparity, they are capable to handle. In this paper a hierarchical approach for disparity estimation is proposed, based on a multiresolution image representation. Intermediate disparity maps are estimated using a com bined FFT-based estimation algorithm in different resolu tion layers. The information from one layer is used as an indicator for the estimation in the successive one. Addi tionally, calculation parameters are adapted to each layer reducing the computational load. The final accumulation with sub-pixel accuracy yields good results even for com plicated images, which can not be reached by other single layer approaches.

Important Links:

Go Back