Image audit

Overview

The image review service is an intelligent review solution from Baidu AI Cloud, offering multidimensional review capabilities, including pornography detection, violent terrorism recognition, political sensitivity analysis, disgusting content identification, and advertisement recognition. It is applicable in various scenarios, such as gaming, social platforms, forums, lifestyle services, and UGC (user-generated content). For example, game developers can use this service to review custom avatars uploaded by players.

The image review service is seamlessly integrated with BOS, allowing you to access its capabilities through BOS API, SDK, CLI, and other methods.

Charge type

As the call entry of the image review service, BOS only charges for BOS API requests. For details, please refer to [BOS Product Pricing](BOS/Product pricing/Product price/Pay-As-You-Go Charge Type.md).

The fee for the image review service is charged by the image review product. For details, please refer to Image Review Product Documentation.

Enable the image review service

Currently, Baidu AI Cloud supports reviewing images on BOS, including actively calling the image review API and automatic image review. For the automatic image review function, you need to first enable the image review service through the console. After the image review rules take effect, the files you newly upload will be automatically reviewed. You can upload objects to BOS through [console](BOS/Console Operation Guide/Manage object/Upload files.md), [SDK](BOS/Developer Guide/Object Basic Operations/Uploading Data/Simple upload.md), CLI and other methods.

1. To enable the image review service on the console, please refer to Setting Up Image Review Service on Console.
2. After enabling the image review service, users can call the image review function through the BOS CLI Tool or BOS API. The following describes the API in detail.

API

This API enables access to the image review service.

Note: To use this API, you need to enable the image review service on the console first.

Request

• Request syntax

  Plain Text Copy

  1POST <ObjectName>?process HTTP/1.1
  2Host: <BucketName>.bj.bcebos.com
  3Date: <Date>
  4Authorization: <AuthorizationString>
  5Content-Type: application/json; charset=utf-8
  6Content-Length: <ContentLength>
  7
  8{
  9	"action" : {
  10		"sync" : [{
  11				"url" : "$(img-censor)",
  12				"parameters" : "<base64_encode(param)>",
  13
  14			}
  15		]
  16	}
  17}

• Request parameters

  • url: The fixed value is $(img-censor), which should not be modified.

  • Parameters: These are defined by the base64-encoded image review function parameters, which are represented as a JSON string.

    Basic structure of image review function parameters:

    Plain Text Copy

    1 {
    2   "antiporn" : {},
    3   "terror" : {},
    4   "ocr" : {
    5       "detect_direction" : "false",
    6       "language_type" : "CHN_ENG",
    7       "recognize_granularity" : "big"
    8   }
    9 }

    May be an arbitrary combination of multiple sub-services as needed, or a single sub-service such as "pornographic content detection", in which case the parameter is written as

    Plain Text Copy

    1{
    2    "antiporn" : {}
    3}

    The list of sub-services supported by the image review service includes:

    • ocr: General optical character recognition
    • face: Face detection
    • antiporn: Pornography recognition
    • politician: Politician recognition
    • terror: Violent terrorism recognition
    • public: Public figure recognition
    • disgust: Disgusting image recognition
    • watermark: Watermark QR code recognition
    • Quality: Image quality recognition values correspond to underlying service parameters. Refer to the input parameters of the Hetu general optical character recognition (OCR) service.

    Detailed explanations of request and response parameters for sub-services are provided below.

• Request headers

  None

Response

• Response headers None

• Response element

  Refer to the explanations for response parameters of each sub-service.

Example

• Request example

  Plain Text Copy

  1{
  2    "antiporn": {},
  3    "ocr": {
  4        "detect_direction": "false",
  5        "language_type": "CHN_ENG",
  6        "recognize_granularity": "big"
  7    }
  8}

After base64 encoding

1eyJhbnRpcG9ybiI6e30sIm9jciI6eyJkZXRlY3RfZGlyZWN0aW9uIjoiZmFsc2UiLCJsYW5ndWFnZV90eXBlIjoiQ0hOX0VORyIsInJlY29nbml6ZV9ncmFudWxhcml0eSI6ImJpZyJ9fQ==

Fill into parameters. The requests sent are as follows:

1POST <ObjectName>?process HTTP/1.1
2Host: <BucketName>.bj.bcebos.com
3Date: <Date>
4Authorization: <AuthorizationString>
5Content-Type: application/json; charset=utf-8
6Content-Length: <ContentLength>
7
8{
9	"action" : {
10		"sync" : [{
11				"url" : "$(img-censor)",
12				"parameters" : "eyJhbnRpcG9ybiI6e30sIm9jciI6eyJkZXRlY3RfZGlyZWN0aW9uIjoiZmFsc2UiLCJsYW5ndWFnZV90eXBlIjoiQ0hOX0VORyIsInJlY29nbml6ZV9ncmFudWxhcml0eSI6ImJpZyJ9fQ==",
13			}
14		]
15	}
16}

• Successful response example

  Plain Text Copy

  1HTTP/1.1 200 OK
  2Date: Thu, 22 Jun 2017 07:30:56 GMT
  3Content-Type: application/json; charset=utf-8
  4Content-Length: 237
  5Connection: keep-alive
  6Server: BceBos
  7x-bce-debug-id: MTAuNzUuNzguNDA6VGh1LCAyMiBKdW4gMjAxNyAxNTozMDo1NiBDU1Q6MTg1MDY5NDg3OQ==
  8x-bce-request-id: 598f7e18-77fb-424a-bc68-95acb0644076
  9{
  10	"result" : {
  11		"antiporn" : {
  12			"result" : [{
  13					"probability" : 0.000071,
  14 "class_name": "Pornography"
  15				}, {
  16					"probability" : 0.000291,
  17 "class_name": "Sexy"
  18				}, {
  19					"probability" : 0.999638,
  20 "class_name": "Normal"
  21				}
  22			],
  23			"log_id" : 1853911322,
  24			"result_num" : 3
  25		},
  26		"ocr" : {
  27			"log_id" : 2471272194,
  28			"words_result_num" : 2,
  29			"words_result" : [{
  30					"words" : " TSINGTAO"
  31				}, {
  32 "words": "Qingdao"
  33				}
  34			]
  35		}
  36	},
  37	"log_id" : 149811665151162
  38}

• Failed response example

  Plain Text Copy

  1{
  2    "log_id": 149319909347709,
  3    "error_code": 216500,
  4    "error_msg": "unknown error"
  5}

  Explanation on failed error_code:

  Error code	Error message	Description
  216101	not enough param	Insufficient parameters
  216102	service not support	An unsupported underlying service type is entered
  216200	empty imge	No image URL
  216500	unknown error	Unknown error
  282804	download image error	Image download failed
  282000	logic internal error	Service logic layer error

ocr

Function

The user sends a request for the service to recognize all characters within an image.

Request parameters

Response parameters

Response example

1{
2"log_id": 2471272194,
3"words_result_num": 2,
4"words_result":
5    [
6        {"words": " TSINGTAO"},
7 {"words": "Tsingtao Beer"}
8    ]
9}

face

Function

• Detects faces in the input image and provides face locations, 72 key point coordinates, and face-related attribute information.
• The detection response time depends on the number of faces in the image. A larger number of faces may slightly slow down the response.
• Typical application scenarios: e.g., Facial Attribute Analysis, Processing Analysis Based on Key Face Points, Facial Marketing Campaigns, etc.
• Specific coordinates are provided for the five sense organs, while the 72 key point coordinates include exact positions but exclude detailed descriptions of their locations.

Request parameters

None

Response parameters

Response example

1{
2    "result_num": 1,
3    "result": [
4        {
5            "location": {
6                "left": 90,
7                "top": 92,
8                "width": 111,
9                "height": 99
10            },
11            "face_probability": 1,
12            "rotation_angle": 6,
13            "yaw": 11.61234664917,
14            "pitch": -0.30852827429771,
15            "roll": 8.8044967651367,
16            "landmark": [
17                {
18                    "x": 105,
19                    "y": 110
20                },
21              ...
22            ],
23            "landmark72": [
24                {
25                    "x": 88,
26                    "y": 109
27                },
28               ...
29            ],
30            "gender": "male",
31            "gender_probability": 0.99358034133911,
32            "glasses": 0,
33            "glasses_probability": 0.99991309642792,
34            "race": "yellow",
35            "race_probability": 0.99960690736771,
36            "qualities": {
37                "occlusion": {
38                    "left_eye": 0.000085282314103097,
39                    "right_eye": 0.00001094374601962,
40                    "nose": 3.2677664307812e-7,
41                    "mouth": 2.6582130940866e-10,
42                    "left_cheek": 8.752236624332e-8,
43                    "right_cheek": 1.0212766454742e-7,
44                    "chin": 4.2632994357028e-10
45                },
46                "blur": 4.5613666312237e-41,
47                "illumination": 0,
48                "completeness": 0,
49                "type": {
50                    "human": 0.98398965597153,
51                    "cartoon": 0.016010366380215
52                }
53            }
54        }
55    ],
56    "log_id": 2418894422
57}

antiporn

Function

Pornography detection.

Request parameters

None

Response parameters

Each element contains the following fields:

Response example

1{
2	"result": [{
3		"probability": 0.000301,
4 "class_name": "Pornography"
5	},
6	{
7		"probability": 0.000054,
8 "class_name": "Sexy"
9	},
10	{
11		"probability": 0.999645,
12 "class_name": "Normal"
13	}],
14 "conclusion": "Normal",
15	"log_id": 848999404,
16 "confidence_coefficient": "Definite",
17	"result_num": 3
18}

politician

Function

Politician identification.

Request parameters

None

Response parameters

Response example

1{
2    "log_id": 3268660173,
3    "result_num": 1,
4    "result": [
5        {
6            "location": {
7                "left": 132,
8                "top": 168,
9                "width": 238,
10                "height": 223
11            },
12            "stars": [
13                {
14 "name": "Zhang San",
15                    "star_id": "515617",
16                    "probability": 0.9750030040741
17                }
18            ]
19        }
20    ]
21}

terror

Function

Violent terrorism recognition

Request parameters

None

Response parameters

Response example

1{
2    "errno": 0,
3    "msg": "success",
4    "data": {
5        "result": 0.0082325544208288,
6        "result_coarse": [
7            {
8 "name": "Normal",
9                "score": 0.99176746606827
10            },
11            {
12 "name": "Violent Terrorism",
13                "score": 0.0082325544208288
14            }
15        ],
16        "result_fine": [
17            {
18 "name": "Normal",
19                "score": 0.98908758163452
20            },
21            {
22 "name": "Police Force",
23                "score": 0.0062405453063548
24            },
25            {
26 "name": "Bloody",
27                "score": 0.0009653537417762
28            },
29            {
30 "name": "Corpse",
31                "score": 0.001054480439052
32            },
33            {
34 "name": "Explosion and Fire",
35                "score": 0.00011743687355192
36            },
37            {
38 "name": "Murder",
39                "score": 0.0011699661845341
40            },
41            {
42 "name": "Riot",
43                "score": 0.000021190358893364
44            },
45            {
46 "name": "Terrorist",
47                "score": 0.0010401027975604
48            },
49            {
50 "name": "Military Weapons",
51                "score": 0.00030337597127073
52            }
53        ]
54    }
55}

public

Function

Public figure recognition

Request parameters

Response parameters

Response example

1{
2    "log_id": 3268660173,
3    "result_num": 1,
4    "result": [
5        {
6            "location": {
7                "left": 132,
8                "top": 168,
9                "width": 238,
10                "height": 223
11            },
12            "stars": [
13                {
14 "name": "Zhang San",
15                    "star_id": "515617",
16                    "probability": 0.9750030040741
17                }
18            ]
19        }
20    ]
21}

disgust

Function

Disgusting image recognition

Request parameters

None

Response parameters

Response example

1{
2	"result": 9.2708455667889E-7,
3	"log_id": 2977989308
4}

watermark

Function

Watermark detection

Request parameters

None

Response Value

Each item in the result contains the following fields:

Response example

1{
2	"result": [{
3		"probability": 0.99872654676437,
4		"type": "watermark"
5	},
6	{
7		"probability": 0.98578763008118,
8		"type": "watermark"
9	}],
10	"log_id": 686882979,
11	"result_num": 2
12}

quality

Function

Image quality

Request parameters

None

Response parameters

Response example

1{
2	"result": {
3		"aesthetic": 0.26410301526388,
4		"clarity": 0.28039423624674
5	},
6	"log_id": 93316007
7}