Why Handwritten Input Recognition Is More Accurate Than OCR?

Why Handwritten Input Recognition Is More Accurate Than OCR?

by

Why Handwritten Input Recognition Is More Accurate Than OCR?

OCR (Optical Character Recognition) is a functionality we commonly use, which translates images into text. Nowadays, OCR can accurately recognize printed fonts, but it struggles with handwritten text. Some say that handwritten input methods can accurately recognize handwriting, even messy handwriting. Why can’t OCR reach this level?

Author: Chaconne

Source: Zhihu
1. Handwriting recognition usually contains more information (here referring to online recognition; the high-accuracy handwriting recognition I’ve encountered has all been online recognition), such as stroke order and connected characters. These details may seem simple, but they provide many features that help improve recognition rates.
2. The sample preprocessing for handwriting recognition is relatively easy. Compared to OCR, handwritten text requires fewer preprocessing steps. Typically, handwriting is entered character by character, with a relatively fixed position and minimal rotation. In contrast, with OCR, you cannot predict the arrangement of the text. OCR samples usually come from scans or photographs, and their resolution is clearly inferior to that of handwritten recognition, plus they often contain a lot of noise. Feature extraction is generally pixel-level, so this noise poses a significant challenge for recognition, especially for Chinese, which has a vast and complex character set.
3. Another characteristic of handwriting recognition is that its recognition process involves human intervention. When a character is input, the recognition engine provides several recognition results based on the input, with the highest scoring result being the default. If this result is not what you want, you can still choose from several candidate results. In contrast, with OCR, one input corresponds to one output, and the final result you see is just the highest scoring classification result, with no real-time control over the recognition.

Related Articles

Can ChatGPT pass the literature graduate entrance exam?

Why is there such a large discrepancy between human and machine grading in AI grading?

How strong is AI in writing poetry?

Wang Shiyuan: Linguistic research must have an interdisciplinary perspective.

The concept of human-computer dialogue existed even before computers appeared.

For more, please click the homepage menu “Linguistics → Miscellaneous Collection (Language and Brain, Development of Linguistics)”

Leave a Comment