Limits...
Visual perception of procedural textures: identifying perceptual dimensions and predicting generation models.

Liu J, Dong J, Cai X, Qi L, Chantler M - PLoS ONE (2015)

Bottom Line: The results suggested that existing dimensions in literature cannot accommodate random textures.We therefore utilized isometric feature mapping (Isomap) to establish a three-dimensional perceptual texture space which better explains the features used by humans in texture similarity judgment.Finally, we proposed computational models to map perceptual features to the perceptual texture space, which can suggest a procedural model to produce textures according to user-defined perceptual scales.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science and Technology, Ocean University of China, 238 Songling Road, Qingdao, Shandong, China; Science and Information College, Qingdao Agricultural University, 700 Changcheng Road, Qingdao, Shandong, China.

ABSTRACT
Procedural models are widely used in computer graphics for generating realistic, natural-looking textures. However, these mathematical models are not perceptually meaningful, whereas the users, such as artists and designers, would prefer to make descriptions using intuitive and perceptual characteristics like "repetitive," "directional," "structured," and so on. To make up for this gap, we investigated the perceptual dimensions of textures generated by a collection of procedural models. Two psychophysical experiments were conducted: free-grouping and rating. We applied Hierarchical Cluster Analysis (HCA) and Singular Value Decomposition (SVD) to discover the perceptual features used by the observers in grouping similar textures. The results suggested that existing dimensions in literature cannot accommodate random textures. We therefore utilized isometric feature mapping (Isomap) to establish a three-dimensional perceptual texture space which better explains the features used by humans in texture similarity judgment. Finally, we proposed computational models to map perceptual features to the perceptual texture space, which can suggest a procedural model to produce textures according to user-defined perceptual scales.

No MeSH data available.


Plot of singular values.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4481328&req=5

pone.0130335.g004: Plot of singular values.

Mentions: The matrix Fmod was decomposed into the product of three matrices:FmodT=U*S*VT(1)U was the matrix of the eigenvectors of which represented the characteristics of models; V was the matrix of the eigenvectors of which represented the characteristics of features; and S was a diagonal matrix of the singular values. We reduced the dimensionality of the solution simply by choosing coefficients in the diagonal matrix, ordinarily starting with the largest. Fig 4 shows the importance that each singular value contributed to the information contained in the original matrix. We kept three singular values and ignored smaller ones, i.e., we reduced U and VT to U3 and , having 3 columns and rows respectively. Leaving out the first column of U3 and , we plotted the second and third columns on the same graph. The reason we abandoned the first column was that, for models, it corresponded to the numbers of features for each model; for features, it correlated with number of times that features had been used in all models. Thus, it was not informative for our purposes. Fig 5 shows the space representing both features and models. The advantage of this technique was that it was able to not only identify the clusters of models, but also link the features to models by measuring distances between points.


Visual perception of procedural textures: identifying perceptual dimensions and predicting generation models.

Liu J, Dong J, Cai X, Qi L, Chantler M - PLoS ONE (2015)

Plot of singular values.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4481328&req=5

pone.0130335.g004: Plot of singular values.
Mentions: The matrix Fmod was decomposed into the product of three matrices:FmodT=U*S*VT(1)U was the matrix of the eigenvectors of which represented the characteristics of models; V was the matrix of the eigenvectors of which represented the characteristics of features; and S was a diagonal matrix of the singular values. We reduced the dimensionality of the solution simply by choosing coefficients in the diagonal matrix, ordinarily starting with the largest. Fig 4 shows the importance that each singular value contributed to the information contained in the original matrix. We kept three singular values and ignored smaller ones, i.e., we reduced U and VT to U3 and , having 3 columns and rows respectively. Leaving out the first column of U3 and , we plotted the second and third columns on the same graph. The reason we abandoned the first column was that, for models, it corresponded to the numbers of features for each model; for features, it correlated with number of times that features had been used in all models. Thus, it was not informative for our purposes. Fig 5 shows the space representing both features and models. The advantage of this technique was that it was able to not only identify the clusters of models, but also link the features to models by measuring distances between points.

Bottom Line: The results suggested that existing dimensions in literature cannot accommodate random textures.We therefore utilized isometric feature mapping (Isomap) to establish a three-dimensional perceptual texture space which better explains the features used by humans in texture similarity judgment.Finally, we proposed computational models to map perceptual features to the perceptual texture space, which can suggest a procedural model to produce textures according to user-defined perceptual scales.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science and Technology, Ocean University of China, 238 Songling Road, Qingdao, Shandong, China; Science and Information College, Qingdao Agricultural University, 700 Changcheng Road, Qingdao, Shandong, China.

ABSTRACT
Procedural models are widely used in computer graphics for generating realistic, natural-looking textures. However, these mathematical models are not perceptually meaningful, whereas the users, such as artists and designers, would prefer to make descriptions using intuitive and perceptual characteristics like "repetitive," "directional," "structured," and so on. To make up for this gap, we investigated the perceptual dimensions of textures generated by a collection of procedural models. Two psychophysical experiments were conducted: free-grouping and rating. We applied Hierarchical Cluster Analysis (HCA) and Singular Value Decomposition (SVD) to discover the perceptual features used by the observers in grouping similar textures. The results suggested that existing dimensions in literature cannot accommodate random textures. We therefore utilized isometric feature mapping (Isomap) to establish a three-dimensional perceptual texture space which better explains the features used by humans in texture similarity judgment. Finally, we proposed computational models to map perceptual features to the perceptual texture space, which can suggest a procedural model to produce textures according to user-defined perceptual scales.

No MeSH data available.