Document Text Watermarking Using Encodable Font Variants

The patent applies local quality by making different parts of the text have different functions. Specific characters are selected as 'encodable characters' that carry both their original semantic meaning and additional encoded information through their font variant. This allows the text to maintain its primary readability while locally embedding watermark data in selected characters without altering the overall visual appearance of the document.

The patent uses parameter changes by varying the font variant parameter of encodable characters to encode information. Each encodable character can be rendered in different font variants (e.g., different weights, italics, underlines) that are imperceptible to human readers but can be detected by OCR software. This changes the physical parameter of the character representation while maintaining the same visual appearance, enabling information encoding without aesthetic alteration.

The patent replaces the mechanical/grayscale approach with a categorical/discrete approach. Instead of using continuous gray level values that are susceptible to noise, the invention uses discrete font variants of specific characters. The OCR system detects which font variant is present (a categorical classification problem) rather than measuring gray levels (a continuous measurement problem). This substitution fundamentally improves reliability because categorical classification is more robust to printing and digitization noise than gray level measurement.

The patent uses copying by creating multiple versions (font variants) of the same character that are visually identical or nearly identical but have detectable differences. The OCR system is trained to recognize these variants and decode the embedded information. This copying approach allows the watermark to be reproduced faithfully through standard printing and digitization processes without degrading the reliability of the encoded information.