Synthesizing new font glyphs from partial observations
View Patent ↗Techniques are disclosed for the synthesis of a full set of slotted content, based upon only partial observations of the slotted content. With respect to a font, the slots may comprise particular letters or symbols or glyphs in an alphabet. Based upon partial observations of a subset of glyphs from a font, a full set of the glyphs corresponding to the font may be synthesized and may further be ornamented.
1. A method for synthesizing a font, the method comprising:
receiving a partial observation of glyph shapes;
generating a full set of glyph shapes by processing said partial observation using a first network that is trained to synthesize glyph shapes; and,
generating a full set of ornamented glyphs by processing said full set of glyph shapes using a second network that is trained to synthesize glyph ornamentation using a conditional adversarial process.
2. The method according to claim 1 , wherein said partial observation of glyph shapes comprises a black and white mask for a glyph.
3. The method according to claim 1 , wherein said first network is trained using a conditional generative adversarial process.
4. The method according to claim 3 , wherein a pre-training operation is performed on said first network using a loss function comprising a combination of a generative adversarial loss and an L 1 loss.
5. The method according to claim 1 , wherein
said first network is trained using a conditional generative adversarial process;
and
a training operation is performed on said first and second networks in an end-to-end configuration using a combination of a generative adversarial loss function, an L 1 loss function and at least one mean square error (“MSE”) loss function.
6. The method according to claim 1 , wherein said full set of ornamented glyphs is generated simultaneously.
7. A system for synthesizing a font, the system comprising:
a first network that is trained to generate a full set of glyph shapes from a partial observation of glyph shapes;
a second network that is trained using a conditional adversarial process to generate a full set of ornamented glyph shapes by synthetically performing ornamentation of said full set of glyph shapes; and,
a coupling module between said first and second network for performing a reshape and grey-scale repetition function on said full set of glyph shapes generated by said first network.
8. The system according to claim 7 , wherein said partial observation of glyph shapes comprises a black and white mask for a glyph.
9. The system according to claim 7 , wherein said first network is trained using a conditional generative adversarial process.
10. The system according to claim 7 , wherein a pre-training operation is performed on said first network using a loss function comprising a combination of a generative adversarial loss and an L 1 loss.
11. The system according to claim 7 , wherein a training operation is performed on said first and second networks in an end-to-end configuration using a combination of a generative adversarial loss function, an L 1 loss function and at least one mean square error (“MSE”) loss function.
12. The system according to claim 7 , wherein said full set of ornamented glyphs is generated simultaneously.
13. A computer program product including one or more non-transitory machine-readable mediums encoded with instructions that when executed by one or more processors cause a process to be carried out for synthesizing a font, the process comprising:
receiving a partial observation of glyph shapes;
generating a full set of glyph shapes by processing said partial observation using a first network that is trained to synthesize glyph shapes using a conditional generative adversarial process, wherein a pre-training operation is performed on said first network using a loss function comprising a combination of a generative adversarial loss and an L 1 loss; and,
generating a full set of ornamented glyphs by processing said full set of glyph shapes using a second network that is trained to synthesize glyph ornamentation.
14. The computer program product according to claim 13 , wherein said partial observation of glyph shapes comprises a black and white mask for a glyph.
15. The computer program product according to claim 13 , wherein said second network is trained using a conditional adversarial process.
16. The computer program product according to claim 13 , wherein
said second network is trained using a conditional adversarial process; and
a training operation is performed on said first and second networks in an end-to-end configuration using a combination of a generative adversarial loss function, an L 1 loss function and at least one mean square error (“MSE”) loss function.