Class DisparityBlockMatchBestFive<I extends ImageBase<I>,​DI extends ImageGray<DI>>

java.lang.Object
boofcv.alg.disparity.DisparityBlockMatchRowFormat<I,​DI>
boofcv.alg.disparity.DisparityBlockMatchBestFive<I,​DI>
Direct Known Subclasses:
DisparityScoreBMBestFive_F32, DisparityScoreBMBestFive_S32

public abstract class DisparityBlockMatchBestFive<I extends ImageBase<I>,​DI extends ImageGray<DI>>
extends DisparityBlockMatchRowFormat<I,​DI>

Scores the disparity for a point using multiple rectangular regions in an effort to reduce errors at object borders, based off te 5 region algorithm described in [1]. Five overlapping regions are considered and the error at each disparity is the center region plus the two regions with the smallest error. The idea is that only errors for common elements in each image are considered.

Scores for individual subregions that the score is computed from are calculated in the same manor as DisparityBlockMatch. Unlike those classes, a rolling window of scores for rectangular regions is saved. From this rolling window of scores for the smaller rectangles the final score is computed by sampling five regions as specified in [1]. The sum of the two best outer regions are added to the center region.

Since five regions are being sampled to compute the score the overall sample region is larger than the individual sub regions. The border radius is thus twice the radius of the sub-regions.

[1] Heiko Hirschmuller, Peter R. Innocent, and Jon Garibaldi. "Real-Time Correlation-Based Stereo Vision with Reduced Border Errors." Int. J. Comput. Vision 47, 1-3 2002