Optical Character Recognition (OCR) refers to the process of analyzing and recognizing image files of text materials to obtain text and layout information. That is, the text in the image is recognized and returned in the form of text.
OCR application scenarios
- Card document identification : Mainland China, Hong Kong, Macao and Taiwan ID cards, passes, passport identification, card identification, vehicle driver's license identification, driving license identification, license identification, enterprise document identification
- Text information structured video recognition : subtitle recognition and text detection, table;
- Bill type identification : VAT invoice identification, all-electric invoice identification, bank check identification, acceptance bill identification, bank bill identification, logistics express identification;
- Other recognition : QR code recognition, one-dimensional code recognition, license plate recognition, mathematical formula recognition, physical and chemical symbol recognition, music symbol recognition, engineering drawing recognition, flow chart recognition, historical site document recognition, handwriting input recognition;
- In addition to the above listed, there are also audit-related business applications such as text recognition in natural scenes, menu recognition, banner detection and recognition, stamp detection and recognition, advertising image and text recognition, etc.
-
1) Provide universal identification services;
2) Some specific scene recognition services that can provide structured text, such as ID card recognition, can retain the structure of the recognized text. However, these applications still have some obvious shortcomings: 1) General recognition services have high image requirements, usually for scanned documents, requiring the input image to have a clean background, simple fonts and neatly arranged text, and have poor text recognition effects in natural scene images; 2 ) Most of them lack common specific scene text recognition, such as
recognition of card images such as business licenses, bank cards, driver's licenses, etc., and only focus on identifying the text content itself, without layout analysis of specific scenes; 3) Specific scene text recognition, the recognition scene is relatively single , for example, Hanwang OCR only provides ID card recognition in specific scenarios, and there is no functional integration of common scenario recognition; 4) Customized function expansion is not possible; 5) Data security is guaranteed by the manufacturer
Below we deploy on rk3568 based on PPocrv3: please download the specific model at ppocr
After exporting onnx, convert it to RKNN,
Please refer to my previous blog for details on how to transfer. . . . . .
Display of results:
The specific code will be uploaded after sorting it out. . . . The code is so poorly written. . . . 0.0