Solution architecture

Advantages

Compared with version 1.0 of the OSS violation detection feature in Content Moderation, the Function Compute solution provides the following advantages:

• You can configure multiple functions to use different detection settings for different OSS buckets.

• Detection logs are written to the corresponding project in Simple Log Service (SLS). This makes it easier to query and analyze detection results.

• Function Compute provides real-time dynamic scaling based on events and requests. This enables incremental image detection in milliseconds.

• It provides virtual machine-level isolation, which lets you process files in OSS buckets with more flexibility and security.

Billing

When you use Function Compute to perform incremental detection on images in an OSS bucket, the following fees are incurred:

Category	Description
Content Moderation service fees	Fees for calling the enhanced image moderation API. For more information, see Introduction to and billing of the enhanced image moderation service V2.0.
OSS access and data transfer fees	Request fees are incurred when you access files in an OSS bucket. For more information, see Request fees. Calling the moderation API using an OSS internal endpoint in the same region results in internal data transfer. Calling the moderation API using a public endpoint or across regions incurs Internet data transfer costs. For more information, see Data transfer fees.
Function Compute invocation and resource fees	Using Function Compute to trigger OSS events and invoke functions incurs resource and invocation fees. For more information, see Billing overview.

Regions supported by the enhanced image moderation service

Prerequisites

• Activate Alibaba Cloud Content Moderation. For more information, see Activation and billing.

• Activate OSS and create a bucket. For more information, see Activate OSS and Create a bucket in the console.

  This topic uses a bucket named osstest-20 as an example.

• Activate Function Compute. For more information, see Activate Function Compute.

Procedure

1. Log on to the Function Compute console, select a region, and then create a service and a function. For more information, see Manage services and Manage functions.

   When you create the function, select Create with Built-in Runtime. The core field configurations are as follows:

   • In the Basic Settings section, set Handler Type to Event Handler.

   • In the Function Code section, select a Runtime, set Code Upload Method to Sample Code, and select oss-trigger.

   • In the Trigger Configurations section, set Trigger Type to Object Storage Service (OSS), Bucket Name to osstest-20, File Prefix to img, and File Suffix to png. For Triggering Events, select oss:ObjectCreated:PutObject and oss:ObjectCreated:PostObject, and set Role Name to AliyunOSSEventNotificationRole.

     When you create a function for the first time, you must grant the event source permission to access Function Compute as prompted. For more information, see Grant an event source permissions to access Function Compute.

     For more information about OSS trigger events, see OSS trigger overview.

2. Use the function to call the enhanced image moderation service of Content Moderation.

   After the function is created, you are redirected to the function code page. On the editor page, you can edit the sample code to call Content Moderation. For more information about the parameters, see Synchronous detection API for the enhanced image moderation service V2.0.

   For more information about how to edit function code, see Code development overview.

   The following code provides a Python example for the enhanced image moderation service. Before you run the sample code, add your AccessKey ID and AccessKey secret to the config.ini configuration file.

   # -*- coding: utf-8 -*-
   import json, uuid
   import configparser
   from aliyunsdkcore.client import AcsClient
   from aliyunsdkcore.request import CommonRequest
   def handler(event, context):
       config = configparser.ConfigParser()
       config.read("config.ini", encoding="GB18030")
       # The configuration file where the AccessKey ID and AccessKey secret are stored.
       evt = json.loads(event)
       evt = evt['events'][0]
       bucket_name = evt['oss']['bucket']['name']
       object_name = evt['oss']['object']['key']
       
       # A public endpoint. The object must have public-read permissions. Data transfer costs over the Internet are incurred.
       objecturl = "https://" + bucket_name + ".oss-cn-hangzhou.aliyuncs.com/" + object_name
       client = AcsClient(config.get("config", "accessKeyId"), config.get("config", "accessSecret"), 'cn-shanghai')
       request = CommonRequest()
       request.set_read_timeout(6000)  # The read timeout period.
       request.set_connect_timeout(3000)  # The connection timeout period.
       request.set_accept_format('json')
       request.set_method('POST')
       request.set_protocol_type('https')  # https | http
       request.set_domain('green-cip.cn-shanghai.aliyuncs.com')
       request.set_version('2022-03-02')
       request.set_action_name('ImageModeration')
       
       request.add_query_param("Service", "baselineCheck")
       request.add_query_param("ServiceParameters",
                               {
                                   'imageUrl': objecturl,
                                   'dataId': str(uuid.uuid4())
                               })
       
       response = client.do_action_with_exception(request)
       print(str(response, encoding='utf-8'))
       
       response_json = json.loads(response)
       if response_json.get("Code") == 200:
           for item in response_json.get("Data", {}).get("Result", []):
               if item["Label"] == "nonLabel":
                   print("nonLabel No label is hit")
               elif "pornographic" in item["Label"] and int(item["Confidence"]) > 80:
                   print("The image may contain pornographic content")
               elif "political" in item["Label"] and int(item["Confidence"]) > 80:
                   print("The image may contain politically sensitive content")
               elif "violent" in item["Label"] and int(item["Confidence"]) > 80:
                   print("The image may contain violent or terrorist content")
               elif "fraud" in item["Label"] and int(item["Confidence"]) > 80:
                   print("The image may contain risky content")
       else:
           print("Request failed. Error code:", response_json.get("Code"))

   For OSS images in the China (Shanghai) or China (Beijing) region, you can use an OSS internal endpoint to call the moderation service in the same region. Before you use this method, you must use your Alibaba Cloud account to grant permissions on the Cloud Resource Access Authorization page. For more information about the parameters, see Synchronous detection API for the enhanced image moderation service V2.0.

   # -*- coding: utf-8 -*-
   import json, uuid
   import configparser
   from aliyunsdkcore.client import AcsClient
   from aliyunsdkcore.request import CommonRequest
   def handler(event, context):
       config = configparser.ConfigParser()
       config.read("config.ini", encoding="GB18030")
       # The configuration file where the AccessKey ID and AccessKey secret are stored.
       evt = json.loads(event)
       evt = evt['events'][0]
       bucket_name = evt['oss']['bucket']['name']
       object_name = evt['oss']['object']['key']
   
       client = AcsClient(config.get("config", "accessKeyId"), config.get("config", "accessSecret"), 'cn-shanghai')
       request = CommonRequest()
       request.set_read_timeout(6000)  # The read timeout period.
       request.set_connect_timeout(3000)  # The connection timeout period.
       request.set_accept_format('json')
       request.set_method('POST')
       request.set_protocol_type('https')  # https | http
       request.set_domain('green-cip.cn-shanghai.aliyuncs.com')
       request.set_version('2022-03-02')
       request.set_action_name('ImageModeration')
       
       request.add_query_param("Service", "baselineCheck")
       request.add_query_param("ServiceParameters",
                               {
                                   'ossRegionId': 'cn-shanghai',
                                   'ossBucketName': bucket_name,
                                   'ossObjectName': object_name,
                                   'dataId': str(uuid.uuid4())
                               })
       
       response = client.do_action_with_exception(request)
       print(str(response, encoding='utf-8'))
       
       response_json = json.loads(response)
       if response_json.get("Code") == 200:
           for item in response_json.get("Data", {}).get("Result", []):
               if item["Label"] == "nonLabel":
                   print("nonLabel No label is hit")
               elif "pornographic" in item["Label"] and int(item["Confidence"]) > 80:
                   print("The image may contain pornographic content")
               elif "political" in item["Label"] and int(item["Confidence"]) > 80:
                   print("The image may contain politically sensitive content")
               elif "violent" in item["Label"] and int(item["Confidence"]) > 80:
                   print("The image may contain violent or terrorist content")
               elif "fraud" in item["Label"] and int(item["Confidence"]) > 80:
                   print("The image may contain risky content")
       else:
           print("Request failed. Error code:", response_json.get("Code"))

3. Perform incremental image detection.

   When you use the OSS PostObject or PutObject operation to upload an image file, you trigger the function. After the detection is complete, you can view the detection results in Invocation Logs. For more information, see View invocation logs.

   To further process OSS files, such as freezing or deleting them, see Manage files.