Correlate.

The images are correlated using the similarity measure specified by cor_type, this could be any of dotprod, eucidean or mad (mean absolute distance). The search range, block height and width are set up in the cor params dialogue box. In addition to the different similarity measures, the correlation algorithm can be either shear, stretch or fast stretch, the parameters of which can also be set up in the cor params dialogue box. The fast stretch algorithm is more efficient than the standard stretch algorithm but gives less flexibility in terms of output and similarity measure, however, for large values of range it speeds up the experiment consiserably.

correlate params The following parameters can be accessed via the interactive dialogue box labeled Cor params:

Height, Width. Dimensions of correlation blocks.

Stretch, Compress. Defines the limits of the stretch algorithm. Both must be positive or zero, and compress should not be bigger than width/2.

Shear. Defines the range of values for the shear algorithm. This number, must be positive.

EDelta. During the creation of the output disparity, a compatibility criterion exists whereby an edge must be present, to within $\pm Edelta$, when the disparity estimate is projected into the right image.

Range. The disparity search range. The search takes place over $initial \pm range$, where initial is supplied by either the input disparity image or is estimated from the regions of interest.

CtF Scaling The stretch correlation algorithm is run initially on a down sampled version of both stereo images, inorder to home in on the solution which is consistent across scales. This choice menu determines the greatest level of down sampling used.