We see that using direct interpolation, PCA gives very bad results - however, if we employ nearest neighbour search and perform regression like we did in the case of LLE reconstruction, we can achieve significantly better results - as can be seen below. We observe that Isomap preserves the Luminosity of the image very well. However, PCA is slightly better at reproducing details like the white dots and the regions around the joint. Also, there is less granuality in the reconstruction from PCA.

Edges are reproduced better in PCA compared to Isomap. Here we see that Isomap produces reconstructions that have badly jagged edges, especially in the red arm.