Learning painting styles for painterly rendering
AssigneeGoogle Inc.
InventorsHui FANG, Genzhi YE
Life of this patent
Estimating the term…
Abstract
Embodiments relate to automatic rendering of images according to a painting style. A procedural painting system, controlled by one or more parameters is used. Classifiers are trained using machine learning to return a score related to how well an image fits a given painting style. Parameters of the procedural painting system are optimized to maximize the score returned by the classifiers, and an output image is rendered.
Claims18
A computer-implemented method, comprising:
- rendering a base output image with a procedural painting system controlled by one or more parameters, wherein the base output image is based on a received input image;
- calculating a score for the base output image, using a classifier trained to calculate a score based on how an image corresponds to a received painting style;
- determining a plurality of randomized sets of meta-parameters for the procedural painting system, wherein each meta-parameter comprises two or more parameters;
- rendering a plurality of intermediate output images by the procedural painting system, wherein the parameters of the procedural painting system for each intermediate output image in the plurality of intermediate output images are determined based on the plurality of randomized sets of meta-parameters;
- calculating a score for each intermediate output image in the plurality of intermediate output images, using the trained classifier;
- and
- identifying an optimal set of metaparameters based on the calculated score for each intermediate output image.
The method of claim 3, wherein the procedural painting system renders the base output image, the intermediate output image, and the final output image using curvilinear patches.
A system, comprising:
- one or more processors;
- and
- a memory in communication with the one or more processors, the memory containing instructions that cause the processor to:
- receive a base output image with a procedural painting system controlled by one or more parameters, wherein the base output image is based on a received input image;
- calculate a score for the base output image, using a classifier trained to calculate a score based on how an image corresponds to a received painting style;
- determine a plurality of randomized sets of metaparameters for the procedural painting system, wherein each metaparameter comprises two or more parameters;
- render a plurality of intermediate output images by the procedural painting system, wherein the parameters of the procedural painting system for each intermediate output image in the plurality of intermediate output images are determined based on the plurality of randomized sets of metaparameters;
- calculate a score for each intermediate output image in the plurality of intermediate output images, using the trained classifier;
- and
- identify an optimal set of metaparameters based on the calculated score for each intermediate output image.
A computer readable storage medium having instructions stored thereon that, when executed by a processor, cause the processor to perform operations including:
- rendering a base output image with a procedural painting system controlled by one or more parameters, wherein the base output image is based on a received input image;
- calculating a score for the base output image, using a classifier trained to calculate a score based on how an image corresponds to a received painting style;
- determining a plurality of randomized sets of metaparameters for the procedural painting system, wherein each metaparameter comprises two or more parameters;
- rendering a plurality of intermediate output images by the procedural painting system, wherein the parameters of the procedural painting system for each intermediate output image in the plurality of intermediate output images are determined based on the plurality of randomized sets of metaparameters;
- calculating a score for each intermediate output image in the plurality of intermediate output images, using the trained classifier,
- identifying an optimal set of metaparameters based on the calculated score for each intermediate output image.
Dependencies are inferred from the claim text.