At the moment, many stereo matching algorithms (global and local) only use intensity values for the disparity map calculation. However, recent studies show that the usage of color information can improve the robustness and quality of matching algorithms. Especially images with radiometric differences pose a challenge to these algorithms. Examples of radiometric differences are different exposure times or different lightning conditions. This thesis deals with the question whether the usage of color information and radiometric transforms can give better results than calculations based solely on intensity. Therefore, based on ten datasets, five color spaces (Intensities, RGB, AC1C2, I1I2I3, LUV), five radiometric transforms (Mean, Laplacian of Gaussian, Rank, SoftRank, Bilateral Subtraction), four cost calculation functions (sum of absolute differences, Birchfield Tomasi, normalized cross correlation, hierarchical mutual information) and two optimization approaches (local, global) are evaluated. Furthermore the usage of weights in radiometric transforms, normalized cross correlation and aggregation are evaluated. A comparision of two weight calculation schemes (Yoon, Geodesic) is also included. The results of the performed evaluation show that the usage of color, radiometric transforms and weights significantly improves the quality of the generated disparity maps.
R. Gross: "Evaluierung konkurrierender Datenterme für das Stereokorrespondenzproblem"; Supervisor: M. Bleyer, M. Gelautz; Institut für Softwaretechnik und Interaktive Systeme, 2009.
Click into the text area and press Ctrl+A/Ctrl+C or ⌘+A/⌘+C to copy the BibTeX into your clipboard… or download the BibTeX.