More Flexible Text Recognition

Version 12 introduces more accurate text recognition and abilities for returning the recognized text at character, word, line or block level. Additionally, one can access several recognition properties, such as confidence, text location and more.

Start from a road sign image.

Recognize blocks of text in the following image.

Return recognition and properties for each recognized line of text.

Perform recognition at word level.

Highlight the location of each recognized word on the image.

Text recognition starts from locating text in an image. In some settings, like when an image is taken in the wild, text detection fails and therefore text recognition fails. TextRecognize now accepts a Masking option that can specify the location of text to recognize in the image.

The Chinese language is not recognized, since TextRecognize is using $Language. TextRecognize can recognize text in different languages and also when multiple languages are present, if prior knowledge is given.