Update atr.qmd

libereurope · Dec 2, 2024 · 83a68dc · 83a68dc
1 parent 87b0964
commit 83a68dc
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/book/atr.qmd b/book/atr.qmd
@@ -25,7 +25,7 @@ Automatic Text Recognition typically involves multiple stages or elements, which
 2. **Binarisation**: Binarisation converts a grayscale or colour image into a binary image, typically using black for the text and white for the background. This simplifies the image, making it easier for the OCR/HTR system to distinguish between the text and non-text elements. This step is especially important for historical documents, where stains, fading, or other degradation can confuse the OCR/HTR engine.
 
 ![](https://github.com/libereurope/ds-essentials/blob/main/book/images/ATR_1.png?raw=true)
-_Examples of binarisation and its effects on legibility, created by Peter Smith who worked to improve_ [_Chinese HTR_](https://blogs.bl.uk/digital-scholarship/2024/03/handwritten-text-recognition-of-the-dunhuang-manuscripts.html) _processes_]
+_Examples of binarisation and its effects on legibility, created by Peter Smith who worked to improve_ [_Chinese HTR_](https://blogs.bl.uk/digital-scholarship/2024/03/handwritten-text-recognition-of-the-dunhuang-manuscripts.html) _processes_
 
 1. **Layout Analysis**: In layout analysis, the system identifies the structure of the document, distinguishing between various elements such as paragraphs, columns, headings, footnotes, and images. It is essential for documents with complex formatting (e.g. newspapers or tables) to ensure that the text is correctly segmented and processed. This step may also involve detecting text regions in multi-layout documents or distinguishing between handwritten and printed text.
 2. **Text Recognition**: The core of the OCR/HTR process is text recognition, where the system identifies individual characters, words, and sentences within the segmented text regions. Modern OCR/HTR engines use pattern recognition techniques or machine learning algorithms, such as neural networks, to improve accuracy. Some systems also perform language modelling, where the recognised text is checked against a dictionary or corpus to ensure contextual correctness, particularly useful for older languages or scripts.