Multi-Wavelet based Stereo Correspondence Matching using Covariance and Enhanced Normalized Cross Correlation

This paper presents the application of covariance and Enhanced Normalized Cross Correlation (ENCC) algorithm in a multi-wavelet based stereo correspondence matching. A balanced multi-wavelet transform is applied to a pair of stereo images de-correlating each input image into its multiwavelet’s sub-bands. The resulting four multi-wavelet basebands of each view are then used to calculate the disparity vectors between the two views. The four multi-wavelet basebands of each view carry different frequency components of that view. A covariance and ENCC block matching algorithm is then used to determine disparity vectors between the two corresponding low resolution sub-bands of the stereo pair images. The resulting four disparity maps from the four basebands are then combined to generate a dense disparity map for the two stereo images. The Middlebury stereo database is used to generate experimental results. A comparison on the simulation results of the proposed technique and state of the arts algorithms indicates the superiority of the proposed technique.