tencent cloud

Feature Overview

The image moderation API supports both sync and async GET request methods. You can use this API to perform content moderation on an image file.

The API supports the following operations:

Note：

• You can moderate image files stored in COS.
• You can moderate images at URLs of a third-party cloud storage vendor.
• Local image files can be moderated after being Base64-encoded by calling the API for batch image moderation.

• Detect COS images or links and recognize content that may be offensive, unsafe, or inappropriate based on the deep learning technology.
• Moderate GIF images by capturing frames.
• Recognize various non-compliant scenarios, including vulgarity, law or regulation violations, pornography, and advertising.
• Detect multiple types of objects (such as object, advertising logo, and QR code) and recognize text in images based on OCR.
  
• Customize moderation policies based on different business scenarios as instructed in Setting Moderation Policy.
• Customize risk libraries to filter non-compliant custom content.

Billing Details

• Each moderation scene is billed separately. For example, if you choose to moderate one image file in two scenes involving pornography and advertising, you will be charged twice.
• Calling the API will incur image moderation fees and COS read request fees as described in Request Fees.
• If the image files are stored in COS STANDARD_IA storage class, calling the moderation API will incur STANDARD_IA data retrieval fees as described in Data Retrieval Fees.
• Image moderation is not supported for objects stored in the ARCHIVE or DEEP ARCHIVE storage classes. To moderate these objects, you first need to restore them as instructed in POST Object restore.
• Moderating audios at URLs of a third-party cloud storage vendor will incur downstream traffic fees charged by the vendor.

Restrictions

• Supported image file size: < 32 MB. To moderate images larger than 5 MB in size, you need to use the large-image-detect parameter when calling the API.
• Supported image file resolution: a resolution of 256x256 or higher is recommended; otherwise, the recognition effect may be affected.
• Supported image file format: PNG, JPG, JPEG, BMP, GIF, or WEBP.
• Supported image URL transfer protocols: HTTP and HTTPS.
• Calling the API requires a signature. For more information, see Request Signature.

SDK Recommendation

COS SDK provides complete capabilities of demo, automatic integration, and signature calculation. You can easily and quickly call APIs through the SDK. For more information, see SDK Overview.

Request

Sample request

GET /<ObjectKey>?ci-process=sensitive-content-recognition&detect-url=<detect-url>&interval=<interval>&max-frames=<max-frames>&biz-type=<biz-type> HTTP/1.1
Host: <BucketName-APPID>.cos.<Region>.myqcloud.com
Date: GMT Date
Authorization: Auth String

Note：

• Authorization: Auth String (see Request Signature for more information)
• When this feature is used by a sub-account, relevant permissions must be granted as instructed in Authorization Granularity Details.

Request headers

This API only uses common request headers. For more information, see Common Request Headers.

Request parameters

The parameters are as described below:

Parameter	Description	Type	Required
ObjectKey	Name of the image file in the COS bucket. The COS bucket is specified by Host. For example, if the file is img.jpg in the test directory in the examplebucket-1250000000 bucket in Beijing, then Host is examplebucket-1250000000.cos.ap-beijing.myqcloud.com, and ObjectKey is test/img.jpg.	String	No
ci-process	This field identifies the data processing feature, which is sensitive-content-recognition for content moderation.	String	Yes
biz-type	Unique identifier of the moderation policy. You can configure the scenes you want to moderate on the moderation policy page in the console, such as pornographic, advertising, and illegal information. For configuration guidelines, see Setting Public Moderation Policy. You can get BizType in the console. If BizType is specified, the moderation request will perform moderation based on the scenes configured in the moderation policy. If BizType is not specified, the default moderation policy will be used automatically.	String	No
detect-url	You can enter a detect-url value to moderate an image accessible over the public network. If detect-url is not specified, the backend will moderate by ObjectKey by default. If detect-url is specified, the backend will moderate by detect-url, and there is no need to enter ObjectKey. Sample detect-url: http://www.example.com/abc.jpg	String	No
interval	For GIF image moderation, you can use this parameter to configure the frame capturing interval. The default value is 5, indicating to capture a frame every five frames starting from the first frame (included).	Int	No
max-frames	The maximum number of frames to be captured for GIF image moderation, which must be greater than 0. The default value is 5, indicating to capture five frames at most.	Int	No
large-image-detect	Whether to compress the image that exceeds the size limit before moderation. Valid values: 0 (no), 1 (yes). Default value: 0. Note: Images up to 32 MB in size can be compressed, and image compression fees will be charged. For large animated images such as GIF, the compression time is long, which may cause the moderation to fail due to timeout.	Int	No
dataid	Image ID. This field will return the original content in the result, which can contain up to 512 bytes.	String	No
async	Whether to moderate asynchronously. Valid values: 0 (returns the result synchronously); 1 (moderates asynchronously). Default value: 0.	Int	No
callback	Callback address. The moderation result (in Detail mode) can be sent to your callback address in the form of a callback. This parameter takes effect for async moderation. Addresses starting with http:// or https:// are supported, such as http://www.callback.com.	String	No

Note：

• Moderation through ObjectKey is a private network operation and will not generate public network traffic.
• Moderation through detect-url will generate public network traffic with regard to the origin where the image resides.

Request body

The request body of this request is empty.

Response

Response headers

This API only returns Common Response Headers.

Response body

<RecognitionResult>
      <JobId>xxxxxxxxxxxxxxx</JobId>
      <Result>1</Result>
      <Label>Porn</Label>
      <SubLabel>SexBehavior</SubLabel>
      <Score>90</Score>
      <PornInfo>
          <Code>0</Code>
          <Msg>OK</Msg>
          <HitFlag>1</HitFlag>
          <Label>xxx</Label>
          <SubLabel>SexBehavior</SubLabel>
          <Score>100</Score>
          <OcrResults>
            <Text></Text>
            <Keywords></Keywords>
            <Location>
              <X></X>
              <Y></Y>
              <Width></Width>
              <Height></Height>
              <Rotate></Rotate>
            </Location>
          </OcrResults>
      </PornInfo>
</RecognitionResult>

The response body is as described below:

Parameter	Type	Description
RecognitionResult	Container	Image moderation result.

The async moderation result RecognitionResult has the following sub-nodes:

Parameter	Type	Description
DataId	String	Image ID. The original content will be returned in the moderation result, which can contain up to 512 bytes.
JobId	String	Image moderation job ID.
State	String	Status of the moderation job. Valid value: Submitted.
Object	String	The name of the image file stored in the COS bucket, which will be returned if ObjectKey is selected during job creation.
Url	String	The URL of the image file, which will be returned if detect-url is selected during job creation.

The sync moderation result RecognitionResult has the following sub-nodes:

Parameter	Type	Description
DataId	String	Image ID. The original content will be returned in the moderation result, which can contain up to 512 bytes.
JobId	String	Image moderation job ID.
State	String	Status of the moderation job. Valid value: Success.
Object	String	The name of the image file stored in the COS bucket, which will be returned if ObjectKey is selected during job creation.
Url	String	The URL of the image file, which will be returned if detect-url is selected during job creation.
CompressionResult	Int	Whether the image is compressed. Valid values: 0 (no), 1 (yes).
Result	Int	This field indicates the moderation result. You can perform subsequent operations based on the result. We recommend you handle different results based on your business needs. Valid values: 0 (normal), 1 (sensitive), 2 (suspiciously sensitive, with human review recommended).
Label	String	This field is used to return the maliciousness label with the highest priority in the detection result, which represents the moderation result suggested by the model. We recommend you handle different types of violations and suggestions based on your business needs. Return values: Normal, Porn, Ads, and other types of unsafe or inappropriate content.
Category	String	This field is a subset of Label, indicating the specific moderation category hit; for example, Sexy represents the sexy category in the Porn label.
SubLabel	String	Second-level label hit by the image.
Score	Int	The confidence the moderation result hits the moderation scene. Value range: 0–100. The higher the value, the more likely the content hits the currently returned moderation scene. For example, Porn 99 means that the content is very likely to be pornographic.
Text	String	The specific text content recognized by OCR in the image, which will be returned if text content detection is enabled in the moderation policy.
PornInfo	Container	The moderation result of the pornographic information moderation scene.
AdsInfo	Container	The moderation result of the advertising information moderation scene.

The moderation information (PornInfo and AdsInfo) has the following sub-nodes:

Parameter	Type	Description
Code	Int	Error code. 0 indicates a success, while other numbers correspond to different errors. For more information, see Error Codes.
Msg	String	The specific error message, which will be OK if the moderation result is normal.
HitFlag	Int	The moderation result returned for the moderation scene. Return values: 0 (normal); 1 (confirmed as a violation of the current scene); 2 (suspected as a violation of the current scene).
Score	Int	The confidence the moderation result hits the moderation scene. Value range: 0–100. The higher the value, the more likely the content hits the currently returned moderation scene. A value in the range of 0–60, 61–90, or 91–100 means the image is normal, suspiciously sensitive, or sensitive respectively. For example, Porn 99 means that the content is very likely to be pornographic.
Label	String	This field indicates the overall result label of the screenshot, which may be SubLabel, a person name, etc.
Category	String	This field is a subset of Label, indicating the specific moderation category hit; for example, Sexy represents the sexy category in the Porn label.
SubLabel	String	This field indicates the specific sub-label hit by the moderation job; for example, SexBehavior is a sub-label under the Porn label. Note: This field may return null, indicating that no specific sub-labels are hit.
OcrResults	Container Array	This field represents the detailed OCR result, including the text recognition result and hit keyword. It will be returned if there is non-compliant content.
LibResults	Container Array	This field returns results based on recognition against the risk library. Note: This field will not be returned if no samples in the risk library are hit.

LibResults has the following sub-nodes:

Parameter	Type	Description
ImageId	String	This field represents the hit image sample ID in the risk library.
Score	Integer	This field returns the confidence under the current label. Value range: 0–100. The higher the value, the more likely the image hits a sample in the risk library. For example, Porn 99 means that the content is very likely to hit a pornographic sample in the library.

OcrResults has the following sub-nodes:

Parameter	Type	Description
Text	String	The specific text content recognized by OCR in the image.
Keywords	String Array	Keywords hit by the current moderation scene.
Location	Container	This parameter is used to return the position (X and Y coordinates of the top-left corner, length, width, and rotation angle) of the OCR detection frame in the image for quick location of the recognized text.

Location has the following sub-nodes:

Name	Type	Description
X	Float	This parameter is used to return the pixel position of the abscissa (X) of the top-left corner of the detection frame. It can be combined with other parameters to uniquely determine the size and position of the detection frame.
Y	Float	This parameter is used to return the pixel position of the ordinate of the top-left corner (Y) of the detection frame. It can be combined with other parameters to uniquely determine the size and position of the detection frame.
Width	Float	This parameter is used to return the width of the detection frame (the length starting from the top-left corner and extending to the right on the X axis). It can be combined with other parameters to uniquely determine the size and position of the detection frame.
Height	Float	This parameter is used to return the height of the detection frame (the length starting from the top-left corner and extending down the Y axis). It can be combined with other parameters to uniquely determine the size and position of the detection frame.
Rotate	Float	This parameter is used to return the rotation angle of the detection frame. Valid values: 0–360 (degrees), and the direction is counterclockwise rotation. This parameter can be combined with the X and Y coordinate parameters to uniquely determine the specific position of the detection frame.

ObjectResults has the following sub-nodes:

Name	Type	Description
Name	String	This field is used to return the name of the recognized object, such as person name.
Location	Container	This parameter is used to return the position (X and Y coordinates of the top-left corner, length, width, and rotation angle) of the recognition result in the image for you to quickly locate information.

Examples

Request 1: Sync image moderation

GET /picture.jpg?ci-process=sensitive-content-recognition&interval=0&max-frames=1&biz-type=*** HTTP/1.1
Host: examplebucket-1250000000.cos.ap-chengdu.myqcloud.com
Date: Tue, 03 Apr 2019 09:06:15 GMT
Authorization:XXXXXXXXXXXX

Response 1

HTTP/1.1 200 OK
Content-Type: application/xml
Content-Length: 645
Date: Tue, 03 Apr 2018 09:06:16 GMT
Status: 200 OK
x-cos-request-id: NWFjMzQ0MDZfOTBmYTUwXzZkZV8z****

<RecognitionResult>
      <JobId>xxxxxxxxxxxxxxx</JobId>
      <CompressionResult>0</CompressionResult>
      <Result>1</Result>
      <Label>Porn</Label>
      <SubLabel>SexBehavior</SubLabel>
      <Score>95</Score>
      <PornInfo>
          <Code>0</Code>
          <Msg>OK</Msg>
          <HitFlag>1</HitFlag>
          <Label>xxx</Label>
          <SubLabel>SexBehavior</SubLabel>
          <Score>95</Score>
      </PornInfo>
</RecognitionResult>

Request 2: Async image moderation

GET /picture.jpg?ci-process=sensitive-content-recognition&biz-type=***&async=1&callback=http://www.callback.com HTTP/1.1
Host: examplebucket-1250000000.cos.ap-chengdu.myqcloud.com
Date: Tue, 03 Apr 2019 09:06:15 GMT
Authorization:XXXXXXXXXXXX

Response 2

HTTP/1.1 200 OK
Content-Type: application/xml
Content-Length: 645
Date: Tue, 03 Apr 2018 09:06:16 GMT
Status: 200 OK
x-cos-request-id: NWFjMzQ0MDZfOTBmYTUwXzZkZV8z****

<RecognitionResult>
      <JobId>xxxxxxxxxxxxxxx</JobId>
</RecognitionResult>