tencent cloud

Live Porn Detection
Download PDF
Last updated: 2023-02-27 15:47:12
Live Porn Detection
Last updated: 2023-02-27 15:47:12
Download PDF
To use the porn detection capability of CSS, you need to enable screencapturing. You can use the porn detection feature either by configuring it in the CSS console or by using APIs. This document shows you how to use porn detection APIs.

Must-Knows

If your COS bucket allows public read access and has politically sensitive, pornographic, or other inappropriate content, to avoid the bucket being blocked, please delete the content first.

Enabling Porn Detection

Because the porn detection feature is based on screencapturing, you need to enable screencapturing first.

1. Create a screencapturing template with porn detection enabled

Call CreateLiveSnapshotTemplate, setting PornFlag to 1 to create a screencapturing template with porn detection enabled.

2. Create a screencapturing rule

Call CreateLiveSnapshotRule, setting TemplateId to the ID of the screencapturing template created in step 1 to associate the template with the target AppId, DomainName, AppName, and StreamName.

3. Start live streaming

After you create a screencapturing rule with porn detection enabled, the porn detection feature will be automatically enabled for new streams. If you want to enable porn detection for an ongoing stream, you need to stop and restart the stream.

Getting the Porn Detection Result

After porn detection is enabled, you can configure a registered domain name in the porn detection callback template to receive callbacks of porn detection results.
Note:
By default, callbacks are sent only if suspicious content is detected.

1. Create a porn detection callback template

Call CreateLiveCallbackTemplate, setting PornCensorshipNotifyUrl to your domain name to create a porn detection callback template.

2. Create a porn detection callback rule

Call CreateLiveCallbackRule, setting TemplateId to the ID of the callback template created in step 1 to associate the template with the target AppId, DomainName, and AppName.

3. Get the porn detection result

The CSS backend will send porn detection results to your domain in the form of HTTP POST requests. You can find the results in JSON format in the request body. The type field indicates whether a live stream contains pornographic content.
Note:
The system cannot achieve 100% accuracy. There may be false positives or false negatives. We recommend you review the images suspected of being pornographic using the type field.

The parameters are as follows:

Parameter
Required
Data Type
Description
streamId
No
String
The stream name.
channelId
No
String
The channel ID.
img
Yes
String
The link of the suspicious image.
type
Yes
Array
The value of the label with the highest priority in the detection result. For details, see the description in label.
score
Yes
Array
The confidence score.
ocrMsg
No
String
The OCR result (if any).
suggestion
Yes
String
The suggestion. Valid values:
Block
Review
Pass
label
Yes
String
The label with the highest priority in the detection result (LabelResults). This is the result generated by the model. Please handle different types of violations based on your business needs.
subLabel
Yes
String
The sub-label of the label with the highest priority in the detection result, such as porn - sexual acts. If no sub-labels are hit, this field will be empty.
labelResults
No
The label hit results generated by the category model, including the detection of pornographic content, ads, terrorist content, and politically sensitive content.
Note: This field may return null, indicating that no valid values can be obtained.
objectResults
No
The detection results generated by the object model, including the label name, hit score, coordinates, scenario, and suggestion for suspicious objects, advertising logos, QR codes, etc. For details, see the data structure of ObjectResults.
Note: This field may return null, indicating that no valid values can be obtained.
ocrResults
No
The OCR result, including the text recognized, the text coordinates, and the suggestion. For details, see the data structure of OcrResults.
Note: This field may return null, indicating that no valid values can be obtained.
libResults
No
The detection results generated by the block/allowlist library.
screenshotTime
Yes
Number
The time when the screenshot was taken.
sendTime
Yes
Number
The Unix timestamp when the request was sent.
stream_param
No
String
The push parameter.
app
No
String
The push domain.
appid
No
Number
The application ID.
appname
No
String
The push path.

LabelResult

The label hit result generated by the category model.
Parameter
Type
Description
Scene
String
The scene identified by the model, such as advertising, pornographic, and harmful.
Suggestion
String
The operation suggested by the system for the current label. Please handle different types of violations based on your business needs. Valid values:
Block
Review
Pass
label
String
The label hit.
SubLabel
String
The sub-label.
Score
Integer
The confidence score for the label.
Details
The sub-label hit details.

LabelDetailItem

The sub-label hit details.
Parameter
Type
Description
Id
Integer
The sequence ID.
Name
String
The sub-label.
Score
Integer
The sub-label score. Value range: 0-100.

ObjectResult

The object detection result.
Parameter
Type
Description
Scene
String
The object scene identified, such as QR code, logo, and OCR.
Suggestion
String
The operation suggested by the system for the current label. Please handle different types of violations based on your business needs. Valid values:
Block
Review
Pass
label
String
The label hit.
SubLabel
String
The sub-label.
Score
Integer
The sub-label score. Value range: 0-100
Names
Array of String
The object names.
Details
Array of ObjectDetail
The object detection details.

ObjectDetail

The object detection details. If the scene identified is object, advertising logo, or QR code, this parameter returns the label name, label value, label score, and location information.
Parameter
Type
Description
Id
Integer
The ID of the object identified.
Name
String
The label hit.
Value
String
The value or content of the label hit. For example, if the label is QR code (QrCode), this parameter is the URL of the QR code.
Score
Integer
The score of the label hit. Value range: 0-100. For example, QrCode 99 indicates a high likelihood that the content is a QR code.
Location
The coordinates (of the top-left corner), dimensions, and rotation of the object detection frame.

Location

The location information of the suspicious content.
Parameter
Type
Description
X
Float
The horizontal coordinate of the top-left corner.
Y
Float
The vertical coordinate of the top-left corner.
Width
Float
The width.
Height
Float
The height.
Rotate
Float
The rotation angle of the detection frame.

OcrResult

The OCR result.
Parameter
Type
Description
Scene
String
The scene identified. Default value: OCR.
Suggestion
String
The operation suggested by the system for the label with the highest priority. Please handle different types of violations based on your business needs. Valid values:
Block
Review
Pass
label
String
The label hit.
SubLabel
String
The sub-label.
Score
Integer
The confidence score of the sub-label. Value range: 0-100.
Text
String
The text.
Details
Array of OcrTextDetail
The OCR details.

OcrTextDetail

The OCR details.
Parameter
Type
Description
Text
String
The text recognized (up to 5,000 bytes).
label
String
The label hit.
Keywords
Array of String
The keywords hit under the label.
Score
Integer
The confidence score of the label. Value range: 0-100.
Location
The OCR text coordinates.

LibResult

The result generated by the block/allowlist library.
Parameter
Type
Description
Scene
String
The scene identified. Default value: Similar.
Suggestion
String
The operation suggested by the system. Please handle different types of violations based on your business needs. Returned values:
Block
Review
Pass
label
String
The label hit.
SubLabel
String
The sub-label.
Score
Integer
The confidence score. Value range: 0-100.
Details
Array of LibDetail
The block/allowlist library detection details.

LibDetail

The custom library or block/allowlist library detection details.
Parameter
Type
Description
Id
Integer
The sequence ID.
ImageId
String
The image ID.
label
String
The label hit.
Tag
String
A custom label.
Score
Integer
The confidence score. Value range: 0-100.

Sample callback

{
        "ocrMsg": "",
        "type": [1],
        "socre": 99,
        "screenshotTime": 1610640000,
        "level": 0,
        "img": "http://1.1.1.1/download/porn/test.jpg",
        "abductionRisk": [],
        "faceDetails": [],
        "sendTime": 1615859827,
        "suggestion": "Block",
        "label": "Porn",
        "subLabel": "PornHigh",
        "labelResults": [{
                "HitFlag": 0,
                "Scene": "Illegal",
                "Suggestion": "Pass",
                "Label": "Normal",
                "SubLabel": "",
                "Score": 0,
                "Details": []
        }, {
                "HitFlag": 1,
                "Scene": "Porn",
                "Suggestion": "Block",
                "Label": "Porn",
                "SubLabel": "PornHigh",
                "Score": 99,
                "Details": [{
                        "Id": 0,
                        "Name": "PornHigh",
                        "Score": 99
                }, {
                        "Id": 1,
                        "Name": "WomenChest",
                        "Score": 99
                }]
        }, {
                "HitFlag": 0,
                "Scene": "Sexy",
                "Suggestion": "Pass",
                "Label": "Normal",
                "SubLabel": "",
                "Score": 0,
                "Details": []
        }, {
                "HitFlag": 0,
                "Scene": "Terror",
                "Suggestion": "Pass",
                "Label": "Normal",
                "SubLabel": "",
                "Score": 0,
                "Details": []
        }],
        "objectResults": [{
                "HitFlag": 0,
                "Scene": "QrCode",
                "Suggestion": "Pass",
                "Label": "Normal",
                "SubLabel": "",
                "Score": 0,
                "Names": [],
                "Details": []
        }, {
                "HitFlag": 0,
                "Scene": "MapRecognition",
                "Suggestion": "Pass",
                "Label": "Normal",
                "SubLabel": "",
                "Score": 0,
                "Names": [],
                "Details": []
        }, {
                "HitFlag": 0,
                "Scene": "PolityFace",
                "Suggestion": "Pass",
                "Label": "Normal",
                "SubLabel": "",
                "Score": 0,
                "Names": [],
                "Details": []
        }],
        "ocrResults": [{
                "HitFlag": 0,
                "Scene": "OCR",
                "Suggestion": "Pass",
                "Label": "Normal",
                "SubLabel": "",
                "Score": 0,
                "Text": "",
                "Details": []
        }],
        "streamId": "teststream",
        "channelId": "teststream",
        "stream_param": "txSecret=40f38f69f574fd51126c421a3d96c374&txTime=5DEBEC80",
        "app": "5000.myqcloud.com",
        "appname": "live",
        "appid": 10000,
        "event_type": 317,
        "sign": "ac920c3e66**********78cf1b5de2c63",
        "t": 1615860427
}


Disabling Porn Detection

You can disable porn detection by deleting the screencapturing rule or modifying the screencapturing template. The change only takes effect for new streams. If you want to disable porn detection for an ongoing stream, you need to stop and restart the stream.

1. Delete the screencapturing rule

Call DeleteLiveSnapshotRule, passing in the DomainName, AppName, and StreamName bound to the screencapturing template ID to delete the screencapturing rule.

2. Modify the screencapturing template

Call ModifyLiveSnapshotTemplate, setting PornFlag to 0.
Was this page helpful?
You can also Contact Sales or Submit a Ticket for help.
Yes
No

Feedback