Product features
This topic answers frequently asked questions about the features, performance, and system logic of Alibaba Cloud Optical Character Recognition (OCR).
Can OCR provide 100% recognition accuracy?
OCR recognition accuracy depends on the quality of the uploaded image. Errors can occur, so 100% accuracy is not guaranteed. If you have questions about the recognition accuracy of the OCR service, contact us.
What are the requirements for image format and size?
The Alibaba Cloud OCR service requires that a single image does not exceed 10 MB. The longest edge of the image cannot exceed 8192 pixels, and the shortest edge must be at least 15 pixels. If the longest edge exceeds 1024 pixels, the aspect ratio cannot exceed 1:50. For faster response times, you can keep the image size under 1.5 MB and call the API by passing an image URL.
There are no specific requirements for image resolution. However, recognition works best when single characters are between 10 and 50 pixels in size. You can use high-definition images without reflections. If an image is rotated, the algorithm automatically corrects its orientation. The recognition rate is highly dependent on the image quality. For more information about supported file types, see the File formats document.
Can the service recognize photocopies?
The APIs for ID cards, bank cards, and business licenses can determine if an image is a photocopy. However, they cannot verify authenticity or detect digital alterations. For other types of certificates with special requirements, you can join the DingTalk group 35208328 to discuss your needs.
Which APIs can recognize multiple types of images?
Typically, each Alibaba Cloud OCR API can recognize only one image at a time. To recognize multiple types of images, you can consider the following product:
Mixed Invoice Recognition supports scenarios where multiple invoices are present in a single image. The system automatically partitions, classifies, and performs structured recognition. You can try it at the OCR Experience Center.
What types of real estate certificates are supported?
The OCR Real Estate Certificate Recognition feature supports two types of certificates: Real Estate Title Certificates and House Property Ownership Certificates. It can recognize various real estate and property rights certificates from different regions across China. For more details, you can try the demo at Duguang.
What should I do if my QPS is insufficient?
The service provides a default traffic limit of 10 queries per second (QPS) for a single account, which supports up to 10 concurrent requests. You can implement request limiting in your program to avoid receiving many throttling errors (Throttling.User). Note: If your program has a retry mechanism for failed OCR recognition, do not retry when the API returns the Throttling.User error code. Retrying may worsen the throttling issue.
Alibaba Cloud also offers QPS add-on packages to increase your capacity. You can purchase them in the OCR console.
What formats do the OCR APIs support?
The Duguang OCR APIs support image formats such as png, .jpg, .jpeg, .jpe, .bmp, .gif, .tiff, .tif, and .webp. Some APIs also support single-page PDF files that do not exceed 10 MB. For more information, see File formats. To recognize other file formats, contact us.
Why is the OCR service not activated even after I purchased a resource plan?
You must activate the pay-as-you-go billing method for OCR before you can use the service. After activation, charges are first deducted from your resource plan. When the resource plan is depleted, the pay-as-you-go billing method is used by default.
Does the OCR service support offline recognition?
OCR supports offline recognition through the sale of software development kits (SDKs). The currently available offline SDKs include those for ID card recognition, bank card recognition, logistics waybill recognition, scan-to-read, and tap-to-read. They are available for purchase at the OCR Alibaba Cloud Marketplace Service Center. If you have other requirements, contact us.
Offline SDKs are not currently supported. This section will be updated with any changes.
Does the OCR service support private deployment?
OCR supports private deployment on Apsara Stack and in hybrid cloud environments. Contact us to discuss private deployment options.
Does the OCR service support international deployment?
The OCR service does not currently support international deployment. If you have this requirement, contact us.
What regions and VPC endpoints does the OCR service support?
The OCR service currently supports access only from the China East 1 (Hangzhou) region. VPC endpoints for other regions, such as Beijing, Kuala Lumpur, and other overseas locations, are not available.
When you use VPC for internal network access, note the following restrictions:
Same-region requirement — The client server must be in the same region as the OCR service (China East 1, Hangzhou). Cross-region access (for example, accessing the Hangzhou VPC endpoint from a Beijing server) is not supported over internal networks.
Cross-region access — If your server is not in the Hangzhou region, you must use a public domain name to access the OCR service.
International deployment — If you call the OCR service from an overseas server (for example, in Kuala Lumpur), the response time may be slow due to cross-border network latency.