Skip to main content

OCR Settings

In this panel, you can not only select the optical character recognition engine and processing model most suitable for your game scene, but also customize image enhancement and paragraph reorganization and merging methods for text blocks through fine adjustment parameters. This allows you to handle various UI interfaces with complex layouts and noise.

Recognition Model and Image Pre-processing Enhancement​

OCR Language Recognition Library​

The system engine provides highly segmented offline visual recognition models to improve processing speed and accuracy. Select the main category through the dropdown menu:

  • Chinese-English-Japanese Model (default): Balances the most mainstream Western and Japanese game interfaces, the main model.
  • Traditional Chinese Model, Korean Model, Latin Alphabet Language Model, Cyrillic Alphabet Language Model, Arabic Model, Greek Model: For cases where only one specific language family appears completely, you can manually switch to avoid possible false matches from the general model.

OCR Speed and Accuracy Model​

Based on your computer's own performance and required response time, you can switch between the following three modes:

  • Standard: Usually recommended. Optimal balance of speed and detection rate.
  • Precise: Recognition speed is slightly slower but text detection rate and punctuation hit rate can achieve extremely high levels.
  • Fast: Extremely low latency for recognition results, suitable for environments running at low resolution or with limited graphics cards, and extremely sensitive to latency.

Image Pre-processing Enhancement​

Game screens are often not the highest quality "black text on white background" layouts. These switches are used to "organize" the image quality before passing the screenshot to the OCR engine:

  • Enhance Contrast: If the game scene is too dark (such as deep in a cave or backlit resulting in dim captured fonts), this can improve the distinction between text brightness and background.
  • Sharpen: When some VR games have too high anti-aliasing settings or insufficient resolution, font edges tend to be blurry. Enable this to reshape and sharpen native text edges.
  • Denoise: Suitable for image quality with heavy grain or lots of interference effects (such as retro filters, pixel art environments, fog particle interference, etc.), can avoid misreading as strange symbols.

OCR Panel Merging and Layout Depth Settings​

Since game interface dialog boxes often have their own wild character spacing and line breaking rules, some line breaks or empty lines will be simply treated as independent sentences resulting in a "car crash scene" for a single translation. All the sliders below (with reset ↺ function) can greatly change the system's merging judgment behavior.

Preset Merging Parameter One-click Strategy​

Pre-set experience value combinations for you. When you arbitrarily pull any of the detailed sliders below, it will also automatically jump to the "Custom" configuration slot.

  • Preset Gear: None (very little merging) / Narrow / Standard / Wide (tolerates more distance breaks caused by messy lines)

Text Block Horizontal Merging Tolerance​

  • Merging Height Tolerance: Limits the overall height difference between two connected text blocks. Through this filter, you can avoid the danger of "UI element impurities" with the same width but different sizes being merged into one sentence. (Default 0.2)
  • Horizontal Height Tolerance: When processing long single-line text, limits the proportional extreme value that can cross the highest and lowest fonts within the horizontal distance. (Default 0.1)
  • Horizontal Overlap Pixel Boundary: Set from -20 to 20px. Negative numbers allow character spacing, positive numbers require mutual overlap or character overlap. (Default -2)
  • Horizontal Y-axis Maximum Spacing: The longitudinal cross-section difference range established to determine whether two horizontally arranged word groups "belong to the same line". (Default 0.5)

Text Paragraph Vertical Reconstruction and Line Break Judgment​

  • Vertical Height Tolerance: Limits the vertical limit tolerance when associating context. (Default 0.3 / 0.5)
  • Vertical X-axis Alignment Tolerance: Limits how many character widths the offset of the first character alignment of two adjacent lines above and below cannot exceed. (Default 1.5)
  • Vertical Maximum Spacing: The empty line truncation limit allowed in the same translation request; exceeding this spacing will be discarded or treated as different sentences. (Default 0.15)
  • Vertical Angle Tolerance: Deals with some dialogue panels with inclined designs or radians. Unit is degrees. (Default 2.0°)
  • Vertical Overlap Minimum Ratio: The minimum基准 allowed when vertically connected elements have上下切边 overlap. (Default 0.0)
  • Vertical Paragraph Spacing Multiplier: When detecting that the vertical distance exceeds how many times the system standard line spacing, the following is认定 as a new paragraph, triggering a forced line break. (Default 1.5)
  • Vertical Minimum New Paragraph Spacing: Same as the above judgment requirement, sets the minimum fixed physical pixel threshold for forced new paragraphs. (Default 80 to 150px)