DocsICP FilingConsole
官方文档
首页

Custom table templates

更新时间：
复制 MD 格式
产品详情
我的收藏

Features

  • Custom table templates are a custom product designed for single-page, fixed-layout, bordered tables and forms. You can use a drag-and-drop interface to configure reference fields, recognition fields, and table headers on a single sample image. This lets you extract structured data from documents with the same layout without data annotation or model training. After configuration and optimization, the template recognition accuracy can exceed 85%.

  • The Toolbox also provides classifier and field type management tools. These tools support automatic classification, routing, and high-precision recognition of data with different layouts through a single API.

Advantages

  • Low cost. You can build a template with just one sample image. No data annotation is required.

  • Easy to use. You can define a template using the drag-and-drop visual configuration. No custom development is needed.

  • High efficiency. You can configure a template in 3 to 5 minutes.

Scenarios

  • Few-shot cold start: This feature is suitable for scenarios with a small number of data samples (fewer than 20) that require fast online validation.

  • Fixed data layout: This feature is suitable for list-type and key-value (KV) type bordered tables with a fixed layout. The style, elements, and table headers must remain unchanged. However, the cell height can be adjusted and rows can be added. Examples include insurance claims processing for medical bills and health check reports, bank transfers, annual report audits, and business data structuring.

    • Note

    • Table templates support both key-value and list-type table data, along with the recognition of fields outside the table borders.

    • The minimum recognition granularity for fields within a table is a single cell. Recognition of areas smaller than a cell is not supported.

    • Recognition of tables or fields that span multiple pages is not supported.

    • For key-value type tables, you can adjust the cell height. Adding, splitting, or deleting cells is not supported.

    • For list-type tables, the table header area must remain unchanged. The number of rows in the recognition area can change. Merged cells are not supported.

Related links

User guide

Watch the following demo to learn how to create a custom table template:

The following figure shows the process of creating a custom table template. You can complete the process in four steps.

image

Step 1: Upload a template image

After you create a custom table template project, go to the project details page to create a template. On the Upload template image page, click the Upload file button to upload a clear, unrotated template image and configure the parameters based on your data's features.

编组 4

Tips

  1. How do I choose a template image?

    1. For the best results, use a scanned image that is clear, properly oriented, free of reflections, and contains all fields and complete table borders.

    2. Documents: Single-page PDF files no larger than 20 MB are supported.

    3. Images: JPG, JPEG, or PNG files no larger than 10 MB are supported.

  2. What are template parameters and how do I configure them?

    1. You can set the feature information for the sample. This helps pre-process the sample data and match it with the AI model to optimize processing.

    2. Configuration items for language, text type, image editing, and red seal processing are supported.

    3. Template parameter settings do not take effect or appear in real time. They take effect during the template testing phase.

  3. What kind of data yields better recognition results?

    1. A single image should have a longest edge of no more than 8192 pixels and a shortest edge of at least 15 pixels. If the longest edge exceeds 1024 pixels, the aspect ratio must not exceed 50:1.

    2. For the best results, keep the size of a single character between 10 and 50 pixels.

    3. Significant differences in the camera angle or orientation between the template image and the actual business data can affect recognition performance.

Step 2: Select reference fields

Go to the Select Reference Fields interface. In the toolbar on the left, select the Box selection button. Draw a box around text or table cells, such as table headers, that have fixed content and positions on the template.

image截屏2023-02-10 10

Reference field: A reference field is a field that is fixed in content and relative position on all documents with the same layout. We recommend that you select reference fields that are distributed around the edges of the page. Within a table, the selection granularity is a single cell. The choice of reference fields also affects performance if you use a joint classifier.

Tips

  1. How do I select reference fields?

    1. Select six or more reference fields. For key-value type data, try to select reference fields that are spread out in the four corners.

    2. For reference fields inside a table, select them at the cell level.

    3. Ensure that the text content and cell position of the selected reference fields are fixed.

    4. Only Chinese, English, and numbers are supported. Symbols or patterns are not allowed.

  2. What if I cannot find six or more reference fields?

    1. You can skip the step of selecting reference fields. The underlying algorithm model performs the analysis automatically. However, if you use a joint classifier for templates with very similar layouts, you must configure them carefully.

Step 3: Configure recognition fields

Go to the recognition field configuration interface. In the left-side toolbar, click the Box selection button. Draw a box around the fields and tables to recognize on the template image. For list-type tables, define the header fields and recognition column fields separately. Click the input box for a header field, enter the corresponding column name, and verify the header value. You can also enter the field name, verify the value, and select the appropriate field type and advanced configuration to improve recognition accuracy.

截屏2023-02-10 10

Recognition field: A recognition field is a cell in the image that needs to be recognized. This is the information that you want to use in your business and is used to output structured results. Each recognition field has required options, such as field name, value, and field type. Each table has required options, such as header fields, recognition column fields, column name, header value, and field type. The advanced configuration is an optional setting.

Field name: The external name of the recognition field. This is the corresponding name in the API operation and must be globally unique.

Value: The content of the recognition field. After you draw a box, the content is automatically recognized and filled in. If you have post-processing requirements for specific fields, such as date normalization, extracting only numbers from amounts, or replacing symbols with regular expressions, you can use field types and advanced configurations. These settings take effect when the API is called.

Column name: The external name of a column in a list-type table, which is the corresponding name in the API operation. Click a header field to set it. The column name must be globally unique.

Header value: The content of the recognition field. After you draw a box, the content is automatically recognized and filled in. If you have post-processing requirements for specific fields, such as date normalization, extracting only numbers from amounts, or replacing symbols with regular expressions, you can use field types and advanced configurations. These settings take effect when the API is called.

Tips

  1. How do I select recognition fields?

    1. Draw a box around the location of the text for each field that you want to extract.

    2. For recognition fields outside a table, draw a box that covers the largest possible area where the text of the field might appear. This helps prevent incomplete field recognition.

    3. For table fields, the minimum granularity is a cell. You should draw the box as close as possible to the border of the cell.

  2. How do I improve the accuracy of field recognition?

    1. Select an appropriate field type to improve recognition results. You can choose a general field type or add a custom field type. If no post-processing is needed, select the regular field type.

    2. To replace content or set a multi-line connector, you can use the advanced configuration.

    3. In the advanced configuration, you can use regular expressions to define the text to be replaced and the replacement text as strings. The following table provides examples:

    Note

    Input

    Original content

    Replacement content

    Result

    Replace string

    abcdef

    abc

    hello

    hellodef

    Remove extra spaces

    ab de f

    \s+

    ab de f

    Add a decimal point before the second to last digit

    12345

    (\d+)(\d{2})$

    \1.\2

    123.45

    Remove content before the colon

    Mobile:12345678

    .+:(\d+)

    \1

    12345678

    Date format conversion

    2023-2-3

    ^(\d+)-(\d+)$

    YYYY-MM-DD

    2023 Year 2 Month 3 Day

Step 4: Test and publish the template

After you select all reference and recognition fields, you can test and publish the template. Click Upload file to upload an image with the same layout to test the template's performance. If you are satisfied with the results, you can publish the template and start making API calls.

Template optimization suggestions

  1. What should I do if the test result shows that the image does not match the template?

    1. Confirm that the uploaded test image has the same layout as the template image.

    2. Adjust the selection area of the reference fields or choose different reference fields to improve the template matching accuracy.

  2. What should I do if the recognition result is incomplete?

    1. Adjust the selection area for the recognition field. The box should cover the largest possible area where the text of the field might appear to resolve this issue.

    2. Check whether the table borders in the test image are complete.

    3. Check the selected field type. For example, if a number is recognized as a letter, you can adjust the field type for that recognition field.

    4. Check the advanced configuration for the field.

  3. What should I do if a recognition field is incorrect in the test results?

    1. Check the selected field type. For example, changing the type from regular to number can lead to more accurate field recognition.

    2. Check whether the selection box for the recognition field aligns with the table cell area.

    3. Check the advanced configuration for the field.

  4. What should I do if the selection box is offset in the test results?

    1. Check whether the selection of reference or recognition fields meets the requirements.

    2. Check whether the template image meets the requirements.

Template management

After the template is published, you can click the project details to go to the project overview. On the project overview page, you can view basic template information, template details, and call data. Published templates can be used through API calls and online use.

Note

Note:

  • You must activate the custom table template service before you can make calls to a published template. Call data is available on the details dashboard. Both API calls and online use are billed. For more information about product billing, see the Billing overview document.

  • After a template is published, you can use the service through online use or API calls.

Previous：Custom KV templatesNext：Form and receipt information extraction