A New Stereo Image Matching Technique using Mutual Information
Clinton Fookes, Arvin Lamanna, Mohammed Bennamoun · UWA Profiles and Research Repository (UWA) · 2001
In this paper we present a new method for matching stereo image pairs. This technique is based on a formulation of mutual information, a new information theoretic topic that has recently seen a prolific expansion in the computer vision and medical imaging field. Mutual information is an intensity-based technique and we formulate this similarity metric with the use of Shannon’s information theorem. Our approach also incorporates adaptive window sizes based on the amount of information contained in the candidate and template windows. Windows containing high information, indicated by large intensity variances, are matched using small square windows. However, windows with low information, indicated by small intensity variances, are matched using larger rectangular windows. Experimental results for a number of stereo pairs indicate that the adaptive mutual information measure shows significant promise for stereo matching, particularly due to its robustness in overcoming the effects of radiometric distortion and improvedmatching in bland regions.