Fig 1.
Example samples in the dataset.
Fig 2.
The similarity matrix constructed from grouping experiments.
The colors indicate the similarity between pairs of samples as specified by the color bar. The labels on the axes represent the 23 procedural models. The distances between the labels represent the number of samples. The green lines separate samples generated by different procedural models. Point colors in block represent the similarity between samples generated by one certain model to another.
Fig 3.
The resulting dendrogram of HCA.
Three clusters below the dissimilarity level of 7 were labeled as Cluster A, Cluster B and Cluster C. Models which were classified as groups below the dissimilarity level of 2.5 were represented by different colors.
Fig 4.
Plot of singular values.
Fig 5.
Plot of texture generation models and features.
The blue stars represent 23 procedural texture generation models and the red circles represent 12 texture perceptual features.
Fig 6.
Plot of residual variance and Isomap dimensions.
Table 1.
Correlation coefficients between the 3 axes (X, Y and Z) of the perceptual space and the average scales of 12 perceptual features for eight subsets of the original data.
Each subset included samples produced by a number of models. Subset 1 to 7 included samples generated by all models except for CA(Forest fire model), Matrix Transformation, texton models, CA models, Cellular, Folding models, folding and fusion models, respectively. Subset 8 contained all samples, i.e. the original data set.
Fig 7.
Magnitude of correlation coefficients between the 3 axes ((A) X, (B) Y and (C) Z) in the perceptual texture space and the average scales of 12 perceptual features for eight subsets.
Fig 8.
Three dimensional representation of the Perceptual Texture Space based on Isomap.
The projection into the (A) x-y plane, (B) x-z plane, (C) y-z plane were shown. Points labeled with yellow, red and blue corresponded to Cluster A, Cluster B and Cluster C resulted by HCA respectively.
Table 2.
Mean squared errors and squared correlation coefficients produced by regression models.
Table 3.
Comparisons of classification accuracy based on using the 12 perceptual features and combinational features learned in the PTS.
Numbers in brackets represented corresponding models that were classified as one group.