Machine Learning Inference of Granular Font Properties
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The imprecision of design metadata font properties and limited availability of font files, such as SVG for proprietary fonts, hinder precise font-based design and analysis, limiting the ability to accurately predict text rendering and perform operations like font substitution and generation.
Innovation Solution
A text analysis system using machine learning techniques to infer granular font properties from rendered text images or design metadata, employing a model trained with synthetic and non-synthetic data, including convolutional neural networks for rotation and scale invariance, and unsupervised pre-training for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If design metadata font properties are used for text rendering, then text can be displayed with specified font size and line height, but the rendering precision is poor and actual values deviate from specified values
Solution Approach 1:
The patent introduces an intermediary machine learning model that takes design metadata font properties as input and predicts actual rendered font properties. This intermediary system bridges the gap between specified font properties and actual rendering results, providing precise predictions of how text will actually appear without requiring access to proprietary font files.
Solution Approach 2:
The patent creates a virtual copy of the font rendering process through machine learning. Instead of accessing actual font files, the system trains a model on rendered text images to learn the mapping from design metadata to actual visual appearance. This copied representation enables precise font property inference without needing the original font data.
2Reliability
If proprietary font files are not available, then font security and intellectual property are protected, but font-based design and analysis operations cannot be performed
Solution Approach 1:
The machine learning model serves as an intermediary that enables font operations without direct access to proprietary font files. The model learns from rendered images and can predict font properties, enable font substitution, and support design operations while maintaining font file security and intellectual property protection.
Solution Approach 2:
The patent replaces the mechanical requirement of accessing font files with a machine learning-based inference system. Instead of directly reading and analyzing font data, the system uses trained models to predict font properties from rendered images, substituting computational inference for direct file access.
3Productivity
If machine learning models are trained with synthetic data, then training efficiency is improved, but the model may lack accuracy on real-world rendered text
Solution Approach 1:
The patent merges synthetic data and real-world data in the training process. Synthetic data provides efficient pre-training with known ground truth, while real-world rendered text images fine-tune the model for actual deployment scenarios. This combination leverages the efficiency of synthetic data generation while ensuring accuracy on real fonts and rendering systems.
Data Source
AI summary
A textual properties model is used to infer values for certain font properties of interest given certain text-related data, such as rendered text images. The model may be used for numerous purposes, such as aiding with document layout, identifying font families that are similar to a given font families, and generating new font families with specific desired properties. In some embodiments, the model is trained from a combination of synthetic data that is labeled with values for the font properties of interest, and partially-labeled data from existing “real-world” documents.