Ten en cuenta que algunos de estos formatos de imagen tienen pérdidas (por ejemplo, JPEG). Si se reduce el tamaño de los archivos de formatos con pérdida, puede que se deteriore la calidad de la imagen y la precisión de los resultados de Document AI.
Resolución de escaneo de documentos
Para obtener los resultados de OCR más precisos de Document AI, los escaneos de documentos deben tener un mínimo de 200 ppp (puntos por pulgada).
En general, las imágenes con una resolución de 300 ppp o superior ofrecen los mejores resultados. La precisión del OCR depende tanto de la resolución como del tamaño de fuente mínimo, así como de otros factores, como la calidad del documento (y de la escritura a mano, si es el caso), por lo que se recomienda hacer pruebas.
La función análisis de calidad de imagen
puede ayudarte a evaluar problemas de resolución.
NOTA: Se necesitan 2000x3000 píxeles para la imagen del reverso del carné de conducir de EE. UU. para poder leer el código de barras.
[[["Es fácil de entender","easyToUnderstand","thumb-up"],["Me ofreció una solución al problema","solvedMyProblem","thumb-up"],["Otro","otherUp","thumb-up"]],[["Es difícil de entender","hardToUnderstand","thumb-down"],["La información o el código de muestra no son correctos","incorrectInformationOrSampleCode","thumb-down"],["Me faltan las muestras o la información que necesito","missingTheInformationSamplesINeed","thumb-down"],["Problema de traducción","translationIssue","thumb-down"],["Otro","otherDown","thumb-down"]],["Última actualización: 2025-09-10 (UTC)."],[[["\u003cp\u003eDocument AI supports various image file types, including PDF, GIF, TIFF, JPEG, PNG, BMP, and WebP, as well as HTML files.\u003c/p\u003e\n"],["\u003cp\u003eFile size and page limits for Document AI are detailed on the Quotas and Limits page.\u003c/p\u003e\n"],["\u003cp\u003eReducing file sizes for lossy image formats like JPEG may result in a degradation of image quality and accuracy of the output from Document AI.\u003c/p\u003e\n"],["\u003cp\u003eFor optimal OCR results, document scans should be at least 200 dpi, with 300 dpi or higher recommended, and US driver's license back side images require 2k x 3k pixels for barcode reading.\u003c/p\u003e\n"],["\u003cp\u003eHTML support is restricted to the Layout Parser feature in Document AI.\u003c/p\u003e\n"]]],[],null,["File Types\n\nDocument AI supports the following image types.\n\nFor information about file size and page limits, refer to the [Quotas and Limits](/document-ai/quotas#content_limits)\npage.\n| **Note:** Document AI includes some supported file types in [Preview](/products#product-launch-stages). These will be charged when they are released to General Availability (GA).\n\n| Name | File Extension(s) | [MIME Type](https://www.iana.org/assignments/media-types/media-types.xhtml) |\n|-------------------------------------------------------------------------------------|-------------------|-----------------------------------------------------------------------------|\n| Portable Document Format (PDF) | `.pdf` | `application/pdf` |\n| Graphics Interchange Format (GIF) | `.gif` | `image/gif` |\n| Tag Image File Format (TIFF) | `.tiff`, `.tif` | `image/tiff` |\n| Joint Photographic Experts Group (JPEG) | `.jpg`, `.jpeg` | `image/jpeg` |\n| Portable Network Graphics (PNG) | `.png` | `image/png` |\n| Bitmap (BMP) | `.bmp` | `image/bmp` |\n| WebP | `.webp` | `image/webp` |\n| HyperText Markup Language (HTML) | `.html` | `text/html` |\n| Microsoft Word Office Open XML (OOXML) ([Preview](/products#product-launch-stages)) | `.docx` | `application/vnd.openxmlformats-officedocument.wordprocessingml.document` |\n| Microsoft PowerPoint OOXML ([Preview](/products#product-launch-stages)) | `.pptx` | `application/vnd.openxmlformats-officedocument.presentationml.presentation` |\n| Microsoft Excel OOXML ([Preview](/products#product-launch-stages)) | `.xlsx` | `application/vnd.openxmlformats-officedocument.spreadsheetml.sheet` |\n\nNote that some of these image formats are \"lossy\" (for example, JPEG). Reducing\nfile sizes for lossy formats may result in a degradation of image quality and\naccuracy of results from Document AI.\n| **Note:** Prior JPEG compressions for TIFF are unsupported. Type of JPEG encapsulation defined by the TIFF [version 6.0 specification](https://gitlab.com/libtiff/libtiff/-/commit/f0a54a4fa0cfa377f493d57ee2af393005d5bbe5).\n| **Note:** HTML and OOXML support are only available with [layout parser](/document-ai/docs/layout-parse-chunk). [Custom splitter](/document-ai/docs/custom-splitter) only supports PDF, TIFF, TIF, and GIF file types.\n\nDocument scan resolution\n\nFor most accurate OCR results from Document AI, document scans should be\na minimum of 200 dpi [(dots per inch)](https://en.wikipedia.org/wiki/Dots_per_inch).\n300 dpi and higher generally produce the best results. OCR accuracy is dependent\non both the resolution and the minimum font size, along with other factors like\ndocument (and if handwritten, handwriting) quality, so testing is recommended.\nThe [image quality analysis](/document-ai/docs/process-documents-ocr#image-quality_analysis)\nfeature can help evaluate resolution concerns.\n\nNOTE: 2k x 3k pixels are required for the US driver's license back side image in\norder to read the barcode."]]
