99 lines
2.3 KiB
Markdown
99 lines
2.3 KiB
Markdown
# AI 图片去重审核系统
|
||
|
||
基于 DashScope 多模态 Embedding + DashVector 向量数据库的图片相似度检测系统。
|
||
|
||
## 功能特性
|
||
|
||
- 使用 DashScope 多模态 Embedding 生成图片向量
|
||
- 使用 DashVector 进行高效向量相似度搜索
|
||
- 支持 pHash 感知哈希预筛选
|
||
- 异步批量下载和处理图片
|
||
- 自动标记重复图片并记录相似度分数
|
||
|
||
## 环境要求
|
||
|
||
- Python 3.8+
|
||
- MySQL 数据库
|
||
- DashScope API Key
|
||
- DashVector API Key
|
||
|
||
## 安装
|
||
|
||
```bash
|
||
pip install -r requirements.txt
|
||
```
|
||
|
||
## 配置
|
||
|
||
创建 `config.ini` 配置文件：
|
||
|
||
```ini
|
||
[database]
|
||
host = localhost
|
||
port = 3306
|
||
user = root
|
||
password = your_password
|
||
database = your_database
|
||
charset = utf8mb4
|
||
|
||
[dashscope]
|
||
api_key = your_dashscope_api_key
|
||
|
||
[dashvector]
|
||
api_key = your_dashvector_api_key
|
||
endpoint = your_endpoint
|
||
collection_name = image_vectors
|
||
vector_dimension = 1024
|
||
|
||
[image]
|
||
cdn_base = https://your-cdn.com/
|
||
|
||
[similarity]
|
||
phash_threshold = 5
|
||
vector_threshold = 0.94
|
||
|
||
[process]
|
||
batch_size = 100
|
||
concurrent_downloads = 10
|
||
log_level = INFO
|
||
log_file = image_similarity.log
|
||
```
|
||
|
||
## 使用方法
|
||
|
||
```bash
|
||
# 处理新图片 (status='draft', similarity='draft')
|
||
python image_similarity_check.py
|
||
|
||
# 重新处理失败的图片 (status='draft', similarity='recalc')
|
||
python image_similarity_recalc.py
|
||
|
||
# 查看统计报告
|
||
python stats_similarity.py
|
||
```
|
||
|
||
## 项目结构
|
||
|
||
```
|
||
├── image_similarity_check.py # 主程序：处理新图片
|
||
├── image_similarity_recalc.py # 重算程序：处理失败的图片
|
||
├── stats_similarity.py # 统计脚本：查看处理结果
|
||
├── query_status.py # 查询处理状态
|
||
├── reset_data.py # 重置数据
|
||
├── reset_vector.py # 重置向量库
|
||
├── config.ini # 配置文件
|
||
└── requirements.txt # 依赖包
|
||
```
|
||
|
||
## 工作流程
|
||
|
||
1. 从数据库获取待处理图片 (`status='draft'`, `similarity='draft'`)
|
||
2. 拼接 CDN URL：`cdn_base + image_url`
|
||
3. 调用 DashScope API 获取 1024 维向量
|
||
4. 在 DashVector 中搜索 topk=3 相似图片
|
||
5. 计算相似度：`similarity = 1.0 - score`
|
||
6. 判断结果：
|
||
- `similarity >= 0.94` → 标记为重复 (`status='similarity'`)
|
||
- `similarity < 0.94` → 标记为不重复 (`status='tag_extension'`)，向量入库
|
||
- 处理失败 → 标记为待重算 (`similarity='recalc'`)
|