---
version: "2026.1.0"
language: "en"
---
# OCR Engine Selection Guide

This guide helps automation engineers choose the right OCR engine and configuration in Leapwork.

## 1. Engine Comparison

Leapwork provides two OCR engines. OCR 3.0 is fast and works best on clean screens, while the Neural Processor is more advanced and handles complex UI scenarios.  

|------------|---------------------------|--------------------|
| Feature    | OCR 3.0                   | Neural Processor   |
| Technology | Classical computer vision | Deep learning      |
| Speed      | Fast                      | Moderate           |
| Best for   | Clean UI                  | Complex UI         |
| Languages  | Leapwork Supported        | Leapwork Supported |
| Offline    | Yes                       | Yes                |
| Processing | Fixed                     | Adaptive           |

![image-20260608-090221.png](https://docs.leapwork.com/__attachments/a_76da10f38048e541bea8527e94f8d81671d6c110a63b3a89040c533c8e3725ab/image-20260608-090221.png?cb=aa69a36172fe635d2ee3c9d0d83d51ee)

## 2. When to Use OCR 3.0

Use OCR 3.0 when your screen is simple, consistent, and high contrast. It is ideal for performance-sensitive automation.  

|-------------|--------------------|
| Scenario    | Example            |
| Plain UI    | Forms, simple apps |
| Reports     | PDF preview        |
| Large text  | Readable UI        |
| High volume | Fast flows         |

## 3. When to Use Neural Processor

Use the Neural Processor when screens are visually complex or OCR 3.0 fails. It provides better accuracy at the cost of speed.  

|------------|------------|
| Scenario   | Example    |
| Dark UI    | Dashboards |
| Gradients  | Headers    |
| Dense data | Tables     |
| Small text | IDs        |

## 4. Presets Overview

### Balanced

Balanced is the default preset and works well for most business applications where text is reasonably sized and clear.  

|-----------|--------------|
| Behavior  | Details      |
| Scale     | 3x           |
| Retry     | Auto         |
| Filtering | Clean output |

### Tiny UI Text

Use this preset when text is small or tightly packed. It improves detection of short values like IDs or codes.  

|------------|----------------|
| Behavior   | Details        |
| Scale      | 5x             |
| Detection  | Sensitive      |
| Confidence | Lower accepted |

### Aggressive Recovery

This preset should be used only when others fail. It maximizes detection but may introduce noise.  

|-------------|-------------------|
| Behavior    | Details           |
| Scale       | 6x                |
| Sensitivity | High              |
| Output      | May include noise |

## 5. Scale Setting

Scale determines how much the image is enlarged before OCR processing. Higher scale improves accuracy for small text but reduces performance.  

|-----------|-------------------|
| Text Size | Recommended Scale |
| Large     | 2--3              |
| Normal    | 3--4              |
| Small     | 5--6              |
| Tiny      | 7--8              |

## 6. Decision Guide

Start with OCR 3.0 for simple screens. If results are incomplete, switch to Neural Processor. Choose preset based on text size and UI complexity.

## 7. Troubleshooting

If OCR results are incorrect, adjust scale, switch presets, or verify language settings.  

|------------------|------------------------|
| Issue            | Solution               |
| Missing text     | Use Neural Processor   |
| Wrong characters | Increase scale         |
| Slow performance | Reduce scale or preset |