There are various factors, such as technology and cost, that limit the performance of OCR services. There are system-wide constraints that affect all services, as well as service-level constraints that affect individual services only.
General Table OCR
- Only images in PNG, JPG, JPEG, BMP, or TIFF format can be recognized.
- No side of the image can be smaller than 15 or larger than 8,192 pixels.
- The area to be recognized must occupy more than 80% of the image. When scanning a table, ensure that the entire table and its surrounding area are included in the image.
- An image can be rotated to any angle.
- Text in images with complex backgrounds (such as outdoor scenery or anti-counterfeit watermarks) or distorted table lines cannot be recognized.
- English and Chinese are both supported, but the support for traditional Chinese characters is limited.
General Text OCR
- Only images in PNG, JPG, JPEG, BMP, GIF, TIFF, WebP, PCX, ICO, PSD, or PDF format can be recognized.
- No side of the image can be smaller than 15 or larger than 8,192 pixels.
- The area to be recognized must occupy more than 80% of the image. When scanning a table, ensure that all text and its surrounding area are included in the image.
- An image can be rotated to any angle.
- Text in images with complex backgrounds (such as outdoor scenery or anti-counterfeit watermarks) or distorted text cannot be recognized.
- Supported languages: Chinese, English, some traditional Chinese, Malay, Ukrainian, Hindi, Russian, Vietnamese, Indonesian, Thai, Arabic, German, Latin, French, Italian, Spanish, Portuguese, Romanian, Polish Amharic, Japanese, Korean, Turkish, Norwegian, Danish, Swedish, Khmer, and Hebrew.
Smart Document Recognizer
- English and Chinese are both supported, but the support for traditional Chinese characters is limited.
- Only images in PNG, JPG, JPEG, BMP, GIF, TIFF, WebP, PCX, ICO or PSD format and PDF files can be recognized. PDF files can only be recognized one page at a time, but you can use the pdf_page_number parameter to specify which page you want to recognize.
- No side of the image can be smaller than 15 or larger than 8,192 pixels.
- The area to be recognized must occupy more than 80% of the image. When scanning a table, ensure that all text and its surrounding area are included in the image.
- An image can be rotated to any angle.
- For more accurate recognition results, the number of characters on a single page must be limited to 1,800 or less.
- Text in images with complex backgrounds (such as outdoor scenery or anti-counterfeit watermarks) or distorted text cannot be analyzed.