Identifying matching fonts utilizing deep learning

1. A non-transitory computer readable medium storing executable instructions which, when executed by a processing device, cause the processing device to perform operations comprising:

generating, utilizing a glyph-based machine learning model, glyph feature vectors from a subset of glyphs of a glyph set, the subset of glyphs being from a digital text depicted within a digital document;

generating, a combined glyph feature vector by combining the glyph feature vectors from the subset of glyphs of the glyph set;

determining, from among a plurality of potential target fonts, one or more target fonts that include matching glyphs corresponding to the subset of glyphs by filtering out one or more potential target fonts from the plurality of potential target fonts that does not include the subset of glyphs of the glyph set;

generating, utilizing the glyph-based machine learning model, a combined target glyph feature vector for each font of the one or more target fonts filtered from the plurality of potential target fonts that include matching glyphs to the subset of glyphs of the glyph set by combining target glyph feature vectors; and

determining, from the combined target glyph feature vectors of the one or more target fonts, a matching font corresponding to the digital text depicted within the digital document by comparing the combined glyph feature vector with the combined target glyph feature vectors.

2. The non-transitory computer readable medium of claim 1 , the operations further comprising: determining probability scores that the combined glyph feature vector matches the combined target glyph feature vector for each font of the one or more target fonts that include matching glyphs.

3. The non-transitory computer readable medium of claim 1 , the operations further comprising:

receiving, from a client device, a selection of the subset of glyphs from a digital text; and

generating, utilizing the glyph-based machine learning model, the glyph feature vectors and the target glyph feature vectors from the one or more target fonts in response to the selection of the subset of glyphs.

4. The non-transitory computer readable medium of claim 1 , the operations further comprising: filtering out the one or more potential target fonts from the plurality of potential target fonts by comparing encoding values of the one or more potential target fonts with encoding values of the subset of glyphs of the glyph set.

5. The non-transitory computer readable medium of claim 1 , the operations further comprising:

determining similarity scores reflecting visual similarity between the glyph feature vectors and the target glyph feature vectors of the one or more target fonts; and

selecting, in relation to the digital text depicted within the digital document, a most visually similar target font as the matching font according to the similarity scores.

6. The non-transitory computer readable medium of claim 1 , the operations further comprising:

generating a glyph image depicting the subset of glyphs;

generating target glyph images depicting respective sets of matching glyphs from the one or more target fonts; and

determining the matching font by comparing the glyph image and the target glyph images.

7. The non-transitory computer readable medium of claim 6 , the operations further comprising comparing the glyph image and the target glyph images by:

generating, utilizing the glyph-based machine learning model, a glyph image feature vector from the glyph image depicting the subset of glyphs from the digital text;

generating, utilizing the glyph-based machine learning model, target glyph image feature vectors from the target glyph images depicting respective sets of matching glyphs from the one or more target fonts; and

determining similarity scores between the glyph image feature vector and the target glyph image feature vectors.

8. A system comprising:

one or more memory devices comprising a glyph-based machine learning model; and

one or more processors configured to cause the system to:

generate, utilizing the glyph-based machine learning model, glyph feature vectors from a subset of glyphs of a glyph set, the subset of glyphs being selected from a digital text;

generate, a combined glyph feature vector by combining the glyph feature vectors from the subset of glyphs of the glyph set;

determine, from among a plurality of potential target fonts, one or more target fonts that include matching glyphs corresponding to the subset of glyphs from the digital text by filtering out one or more potential target fonts from the plurality of potential target fonts that does not include the subset of glyphs of the glyph set;

generate, utilizing the glyph-based machine learning model, a combined target glyph feature vector for each font of the one or more target fonts filtered from the plurality of potential target fonts that include matching glyphs to the subset of glyphs of the glyph set by combining target glyph feature vectors; and

determine a matching font corresponding to the digital text by comparing the combined glyph feature vector from the subset of glyphs of the digital text with the combined target glyph feature vector.

9. The system of claim 8 , wherein comparing the combined glyph feature vector from the subset of glyphs of the digital text with the combined target glyph feature vector for each font of the one or more target fonts that include matching glyphs comprises determining similarity scores reflecting visual similarity between the combined glyph feature vector and the combined target glyph feature vector for each font of the one or more target fonts that include matching glyphs.

10. The system of claim 9 , wherein the one or more processors are configured to cause the system to determine the matching font corresponding to the digital text by selecting a target font with a highest similarity score indicating a highest visual similarity in relation to the digital text.

11. The system of claim 10 , wherein the one or more processors are configured to cause the system to:

filter out, the one or more potential target fonts from the plurality of potential target fonts by comparing encoding values of the one or more potential target fonts with encoding values of the subset of glyphs of the glyph set; and

generate, utilizing the glyph-based machine learning model, the target glyph feature vectors only for target fonts that are not filtered out.

12. The system of claim 8 , wherein the one or more processors are configured to cause the system to:

receive a user selection of the subset of glyphs from the digital text; and

determine the matching font corresponding to the digital text in real time with the selection of the subset of glyphs.

13. The system of claim 12 , wherein the one or more processors are configured to cause the system to determine the matching font in real time with the selection of the subset of glyphs by generating the glyph feature vectors and the target glyph feature vectors utilizing the glyph-based machine learning model in real time with the selection.

14. The system of claim 8 , wherein the one or more processors are configured to cause the system to:

filter out repeat glyphs from the subset of glyphs; and

generate, utilizing the glyph-based machine learning model, the glyph feature vectors from unique glyphs that are not filtered out for repetition.

15. A computer-implemented method comprising:

generating, utilizing a glyph-based machine learning model, glyph feature vectors from a subset of glyphs of a glyph set, the subset of glyphs being from digital text depicted within a digital document;

generate, a combined glyph feature vector by combining the glyph feature vectors from the subset of glyphs of the glyph set;

determining, by filtering out potential target fonts from a plurality of potential target fonts that does not include the subset of glyphs of the glyph set from the digital text, one or more target fonts that include matching glyphs corresponding to the subset of glyphs from the digital text;

generate, utilizing the glyph-based machine learning model, a combined target glyph feature vector for each font of the one or more target fonts filtered from the plurality of potential target fonts that include matching glyphs to the subset of glyphs of the glyph set by combining target glyph feature vectors; and

determining, from the combined target glyph feature vectors of the one or more target fonts, a matching font corresponding to the digital text depicted within the digital document by comparing the combined glyph feature vector with the combined target glyph feature vectors.

16. The computer-implemented method of claim 15 , wherein determining the matching font comprises:

comparing the combined glyph feature vector from the subset of glyphs of the digital text with the combined target glyph feature vector for each font of the one or more target fonts that include matching glyphs; and

selecting the matching font according to comparing the combined glyph feature vector with the combined target glyph feature vector for each font of the one or more target fonts that include matching glyphs.

17. The computer-implemented method of claim 15 , further comprising:

receiving, from a client device, a selection of a subset of glyphs from digital text depicted within a digital document; and

generating, utilizing the glyph-based machine learning model, the glyph feature vectors from the digital text and the target glyph feature vectors from the one or more target fonts in real time with the selection from the client device.

18. The computer-implemented method of claim 15 , further comprising:

filtering out repeat glyphs from the subset of glyphs;

generating, utilizing the glyph-based machine learning model, the glyph feature vectors from unique glyphs that are not filtered out for repetition; and

generating, utilizing the glyph-based machine learning model, the target glyph feature vectors from matching glyphs corresponding to the unique glyphs.

19. The computer-implemented method of claim 15 , wherein determining the matching font comprises:

determining similarity scores reflecting visual similarity between the glyph feature vectors of the digital text and the target glyph feature vectors of the one or more target fonts; and

selecting, in relation to the digital text depicted within the digital document, a most visually similar target font as the matching font according to the similarity scores.

20. The computer-implemented method of claim 15 , further comprising determining the one or more target fonts that include matching glyphs corresponding to the subset of glyphs in real time with a selection of the subset of glyphs.