Dimensionality reduction for cuboids descriptors. Samples cuboids from all datasets, and uses this sample to get the pca coefficient for the descriptor (see imagedesc_getpca). Should be run after features are detected (of course), and before descriptor is applied to all the cuboids. This step is optional, but it is probably a good idea since otherwise descriptor can potentially have very large dimension. INPUTS DATASETS - array of structs, should have the fields: .cuboids - length N cell vector of sets of cuboids cubdesc - cuboid descriptor [see imagedesc_getpca] kpca - number of dimensions to reduce data to [see imagedesc_getpca] OUTPUTS cubdec - cuboid descriptor with pca info [see imagedesc_getpca] cuboids - sampled cuboids See also FEATURESSM, PCA, IMAGEDESC_GETPCA