This document describes how to use the image moderation feature in the console. After you enable image moderation, new images uploaded to a bucket will be automatically moderated, and the identified non-compliant content can be automatically blocked (by denying public read access to the content).
You can also scan historical image files in COS to detect pornographic, illegal, and advertising content. For more information, see Single Image Moderation.
1. Log in to the COS console. On the Bucket List page, click the target bucket to enter the bucket details page.
2. On the left sidebar, select Sensitive Content Moderation > Automatic Moderation Configuration and click Image Moderation.
3. Click Add Automatic Image Moderation Configuration and set the following configuration items:
Moderation Scope: You can select The whole bucket or Specified Range.
Path: If you select Specified Range, enter the path of the images to be moderated.
Example 1: To moderate files in the test directory, set this field to test/.
Example 2: To moderate files prefixed with 123, set this field to 123.
Note:
You can add multiple moderation configurations, but the paths cannot be duplicate or have inclusion relationships. If you have configured to moderate the entire bucket, you cannot add a moderation configuration for a specific path in the bucket.
Moderation Suffix: The following image formats can be moderated: JPG, JPEG, PNG, BMP, WEBP, and GIF. Two options are supported: Smart and Without suffix.
Note:
Smart: After you select this option, the system will intelligently determine whether a file is an image based on its file extension and content.
Large Image Moderation: The image moderation feature can moderate images below 5 MB in size. For larger images, you can enable the large image moderation feature. Then, the backend will compress images (up to 32 MB) first before moderating them.
Moderation Policy: Select a moderation policy. You can create different policies for refined moderation. Moderation scene options include Pornographic, Illegal, and Advertisement, and you can select one or multiple options. For detailed directions on how to configure a moderation policy, see Setting Moderation Policy.
Moderation Scene: It displays the default scene or the scene that you configure in the moderation policy. You can select the target scene as needed.
File block configuration: You can enable this service to authorize CI to perform automatic or human moderation and block the identified non-compliant files by denying public read access to them.
Block mode: The following two block modes are supported:
Change the file ACL to private read: Doing so actually blocks the file. Then, a 403 status code will be returned when the file is accessed again, indicating that access is denied. For more information on file permissions, see ACL.
Transfer the file to the backup directory: Doing so actually blocks the file. Then, a 404 status code will be returned when the file is accessed again, indicating that the file does not exist. The backup directory is automatically generated by the backend at audit_freeze_backup/increment_audit in the current bucket.
Block Type: You can select a block type and mechanism. Machine moderation and block is selected by default. If you select Human moderation and block, Tencent Cloud security team will review suspiciously sensitive images identified during machine moderation. You can select the image score range for blocking (by specifying an integer between 60 and 100; the greater the score, the more sensitive the image).
Callback: After callback is enabled, you will receive image moderation results. You need to select the moderation type, callback content, callback URL, and image domain name. If you select Custom Callback Threshold, you need to set the score range of the images for callback. After the callback URL is set, CI will send the default callback message to the set URL to check whether it can receive callback messages normally. For more information on callback, see Image Moderation Callback Content.
4. After completing the configuration, click Save. Images uploaded subsequently will be moderated.
Notes
1. Image moderation adopts a scoring mechanism, with a score between 0 and 100 returned for each image.
2. The confirmed part refers to the images that are confirmed to be normal or sensitive (with a score in the range of [0,60] or (90,100]). For such images, no manual intervention is required.
3. The uncertain part refers to the images that are suspected to be sensitive (with a score in the range of (60,90]). For such images, human moderation is recommended.
