Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Fig 1.

81 x 110cm sections of the real (left) and replica (right) Blue Poles: Number 11, 1952 (210 x 486.8cm, National Gallery of Australia).

More »

Fig 1 Expand

Table 1.

A selection of AI techniques listed in chronological order.

More »

Table 1 Expand

Table 2.

Data partitioning of Pollock and non-Pollocks according to the number of artworks, the number of images of these artworks, and the number of tiles in these images.

More »

Table 2 Expand

Fig 2.

A flow diagram of the image pre-processing, data partitioning, and machine learning.

The image is cropped and then tiled at multiple size scales (3 of the 22 tile scales are shown). The machine trains on the full set of tiles. During inference, the machine assigns a Pollock Signature to each of the tiles, which are then grouped by size and an average Pollock Signature is assigned. Finally, the average Pollock Signature is averaged across tile sizes to calculate the PMF. The intermediate numbers shown in the diagram are Pollock Signatures of the tiles and the final number is the PMF.

More »

Fig 2 Expand

Fig 3.

Mean MA for the various model architectures considered.

The blue bars represent CNN model architectures, including ResNet50. The orange/red bars represent ViT model architectures. Within each model architecture, the error bars quantify the variability in MA performance between models generated by choosing different random seeds during training. The green bar represents the highest performing model within the ResNet50 model architecture and ultimately becomes our chosen model. All models are trained using our multi-scaled tiling method, except for the red group models which are trained on the full images. While the red bars are trained using a batch size of 4 due to computational limitations, the faded red bar is trained with a batch size of 32.

More »

Fig 3 Expand

Fig 4.

An example of an adult (top) and child (bottom) Dripfest image (61.5 x 91.5 cm).

More »

Fig 4 Expand

Table 3.

Data partitioning of Dripfest and non-Dripfest artworks according to the number of artworks.

More »

Table 3 Expand

Fig 5.

The Pollock Dial (see text for details).

More »

Fig 5 Expand

Fig 6.

A histogram of the number of Pollock (green) and non-Pollock (red) images featured in the Pollock Dial. The number of images, n, in a given PMF range is plotted on the left y axis and the PMF values are plotted on the x axis. The black vertical line represents the chosen PMF Threshold. The black line is the MA obtained by shifting the threshold to different PMF values. These MA values are plotted on the right-hand y axis.

More »

Fig 6 Expand

Fig 7.

A Plot of PMF versus Resolution Fraction (RF) for the 2 sets of Pollock images: blue (high resolution, color images) and orange (lower resolution, grayscale images). The top images show close-ups of Pollock’s Blue Poles and Dummy at RF = 1 (scanned resolution, left) and RF = 1/30 (cliff resolution, right).

More »

Fig 7 Expand

Fig 8.

A plot of PMF versus RF for the Pollock (green) and non-Pollock (red) images on the Pollock Dial.

More »

Fig 8 Expand

Fig 9.

Average PMF plotted against the image contrast (top) and image brightness (bottom) for the Pollock (green) and non-Pollock images (red). The broad colored regions correspond to the standard deviations in the data. The horizontal line represents the PMF Threshold. In each case, 0 corresponds to the undistorted value on the x axis.

More »

Fig 9 Expand

Fig 10.

Zoom-in Chart for Dummy (see text for details).

The images above the Chart show example tile images.

More »

Fig 10 Expand

Fig 11.

The Zoom-In Chart averaged across all non-Pollock poured paintings superimposed on the Zoom-In Chart averaged across all Pollock poured paintings.

More »

Fig 11 Expand

Fig 12.

The Zoom-in Chart for Blue Poles (see text for details).

The images above the Chart show example tile images.

More »

Fig 12 Expand

Fig 13.

Spatial investigation of Dummy’s Signatures.

Grayscale image of Dummy (top left panel) and the same image with the Signature Map overlaid (right panel). The bottom panel focuses on Dummy’s Map. The panel above the Map plots the average Pollock Signature as we move from left to right across the canvas and the panel to the right does the same as we move from bottom to top.

More »

Fig 13 Expand

Fig 14.

Spatial investigation of Blue Poles’s Signatures.

Grayscale image of Blue Poles (top left panel) and the same image with the Signature Map overlaid (right panel). The bottom panel focuses on Blue Poles’s Map. The panel above the Map plots the average Pollock Signature as we move from left to right across the canvas and the panel to the right does the same as we move from bottom to top.

More »

Fig 14 Expand

Fig 15.

Spatial investigation of Pollock’s Untitled: Cut-Out (77.3 x 57cm, Ohara Museum of Art, Japan).

Grayscale image of Cut-out (top left panel) and the same image with the Signature Map overlaid (right panel). The bottom panel focuses on Cut-Out’s Map. The panel above the Map plots the average Pollock Signature as we move from left to right across the canvas and the panel to the right does the same as we move from bottom to top.

More »

Fig 15 Expand

Fig 16.

A timeline chart of the evolution of the various machine parameters (PMF, SI, M, C, and U) along with other artwork parameters (N, A, and AR) spanning from 1943 to 1954.

See text for details.

More »

Fig 16 Expand

Fig 17.

Plots of SI, M, C, and U against PMF (see text for details).

More »

Fig 17 Expand

Fig 18.

Radial maps for Pollock’s Water Birds (66.4 x 53.8cm, Baltimore Museum of Art, USA) (top row), Pollock’s Blue Poles (middle row), and Henri Michaux’s Untitled (74.9 x 107.9cm, Edward Thorp Gallery, New York) (bottom row). The left column shows the artwork, the middle column shows the radial map plotted using the absolute colors, and the right column shows the radial map plotted using relative colors. The top and bottom bars show the relative and absolute color ranges, respectively. See text for details.

More »

Fig 18 Expand

Fig 19.

Left: A 63.5 x 63.5cm section of Pollock’s Number 32, 1950 (269 x 457.5cm, Kunstsammlung Nordrhein-Westfalen, Germany). Middle-left: A 100 x 100cm computer generated fractal image. Middle-right: the Map of the fractal image. Right: the Map constructed from 10cm tiles (this tile size is shown in the middle left image as a yellow square).

More »

Fig 19 Expand

Fig 20.

Plots of PMF versus fractal dimension D for the computer-generated fractals (red) and for Pollocks (blue). The inserts show example fractal images for D = 1.2, 1.5, and 1.8.

More »

Fig 20 Expand

Fig 21.

Comparisons of a section of Pollock’s One: Number 31, 1950 (269.5 x 530.8cm, MoMA, New York) (middle image) to photographs of a thicket (top image) and trees (bottom image).

More »

Fig 21 Expand

Fig 22.

Sample thumbnails of the test patterns.

From left to right (single colors, gradient grayscale images, black and white rows (at various angles), colored rows (at various angles), gradient grayscale images (at various angles), and pickup sticks.

More »

Fig 22 Expand

Fig 23.

Poured painting generated by the wind machine (PMF = 0.11).

More »

Fig 23 Expand

Fig 24.

The Pollockizer (left) and a section of the poured painting that is classified as sharing the visual signatures of a Pollock (PMF = 0.67) (right).

More »

Fig 24 Expand

Fig 25.

A comparison of sections of Henri Michaux’s SansTitre 1960–61 (bottom, PMF = 0.74) with Pollock’s Untitled 1950 (top, PMF = 1.00).

More »

Fig 25 Expand

Fig 26.

Average PMF plotted against the image contrast (top) and image brightness (bottom) for the Pollock (green) and non-Pollock images (red). The broad colored regions correspond to the standard deviations in the data. The horizontal line represents the PMF Threshold. In each case, 0 corresponds to the undistorted value on the x axis.

More »

Fig 26 Expand