阿里旷视百度 OCR图片文字识别功能对比

最新推荐文章于 2025-06-30 14:52:52 发布

wwdwjm

最新推荐文章于 2025-06-30 14:52:52 发布

阅读量1w

点赞数 2

CC 4.0 BY-SA版权

文章标签： orc

本文链接：https://blog.youkuaiyun.com/wwdwjm/article/details/76608101

本文比较了阿里、旷视和百度的OCR图片文字识别功能。结果显示，尽管每个API都有识别不完全的情况，但百度的识别率相对较高。阿里OCR支持异步处理，适合对实时性要求不高的项目。测试中，复杂图片的识别效果普遍较差。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

阿里旷视百度 OCR图片文字识别功能对比

因为项目中需要用到图片转文字的功能看到网上有挺多这方面的API 所以测试下哪个产品的功能更好目前选中了阿里旷视和百度这三家的API

图片：

旷视识别出来的结果

句对起

沨1当初的爱你

下面的是旷视把结果返回的json字符串

{"image_id":"AvOVDQxfja9cu+rL0mYurQ==", "result":[{"child-objects": [{"child-objects": [],"position": [{"y": 319, "x": 123},{"y": 319, "x": 137}, {"y": 333, "x":137}, {"y": 333, "x": 123}], "type":"character", "value": "m"}],"position": [], "type": "textline","value": "m"}, {"child-objects":[{"child-objects": [], "position": [{"y": 134,"x": 42}, {"y": 133, "x": 61}, {"y":153, "x": 63}, {"y": 155, "x": 43}],"type": "character", "value":"\u53e5"}, {"child-objects": [], "position":[{"y": 132, "x": 66}, {"y": 134, "x":86}, {"y": 154, "x": 84}, {"y": 152,"x": 64}], "type": "character","value": "\u5bf9"}, {"child-objects": [],"position": [{"y": 133, "x": 112},{"y": 135, "x": 133}, {"y": 156, "x":131}, {"y": 154, "x": 110}], "type": "character","value": "\u8d77"}], "position": [],"type": "textline", "value":"\u53e5\u5bf9\u8d77"}, {"child-objects": [{"child-objects":[], "position": [{"y": 136, "x": 22},{"y": 137, "x": 34}, {"y": 149, "x":33}, {"y": 149, "x": 21}], "type":"character", "value": "\u4e00"}],"position": [], "type": "textline","value": "\u4e00"}, {"child-objects":[{"child-objects": [], "position": [{"y": 157,"x": 17}, {"y": 156, "x": 39}, {"y":178, "x": 41}, {"y": 180, "x": 19}],"type": "character", "value": "\u6ca8"},{"child-objects": [], "position": [{"y": 158,"x": 55}, {"y": 157, "x": 77}, {"y":178, "x": 78}, {"y": 179, "x": 56}],"type": "character", "value": "1"},{"child-objects": [], "position": [{"y": 157,"x": 88}, {"y": 158, "x": 108}, {"y":179, "x": 108}, {"y": 179, "x": 87}],"type": "character", "value":"\u5f53"}, {"child-objects": [], "position":[{"y": 157, "x": 112}, {"y": 160, "x":133}, {"y": 180, "x": 131}, {"y": 178,"x": 110}], "type": "character","value": "\u521d"}, {"child-objects": [],"position": [{"y": 159, "x": 137},{"y": 158, "x": 156}, {"y": 178, "x":157}, {"y": 179, "x": 137}], "type":"character", "value": "\u7684"},{"child-objects": [], "position": [{"y": 158,"x": 161}, {"y": 158, "x": 177}, {"y":178, "x": 178}, {"y": 179, "x": 161}],"type": "character", "value":"\u7231"}, {"child-objects": [], "position":[{"y": 158, "x": 183}, {"y": 158, "x":203}, {"y": 178, "x": 203}, {"y": 178,"x": 183}], "type": "character","value": "\u4f60"}], "position": [],"type": "textline", "value":"\u6ca81\u5f53\u521d\u7684\u7231\u4f60"}, {"child-objects":[{"child-objects": [], "position": [{"y": 136,"x": 90}, {"y": 135, "x": 106}, {"y":152, "x": 107}, {"y": 153, "x": 91}],"type": "character", "value":"\u4e0d"}], "position": [], "type": "textline","value": "\u4e0d"}], "request_id":"1501724700,03d59c8d-3fd1-447e-88d4-553d9b9814a7","time_used": 771}

阿里识别出来的结果

句对7网当

下面是返回结果的json字符串

{

"msg":"OK",

"code":200,

"data":[

{

"msg":"OK",

"code":200,

"dataId":"763a141f-c6f7-4c4e-a8c5-b3898d1d2d65",

"results":[

{

"rate":98.9,

"suggestion":"pass",

"label":"normal",

"scene":"porn"

{

"ocrData":[

"句对7网当"

"rate":99.91,

"suggestion":"review",

"label":"ocr",

"scene":"ocr"

{

"rate":99.91,

"suggestion":"pass",

"label":"normal",

"scene":"qrcode"

{

"rate":99.91,

"suggestion":"pass",

"label":"normal",

"scene":"sface"

}

"taskId":"imgSqAsQ6LAOf6s4TV0V7Oc2-1nqn0g",

"url":"https://timgsa.baidu.com/timg?image&quality=80&size=b9999_10000&sec=1501734176984&di=3490f64809b7b6aad63248c1781b2297&imgtype=0&src=http%3A%2F%2Fcdn.duitang.com%2Fuploads%2Fitem%2F201405%2F20%2F20140520230352_LSSmd.thumb.224_0.jpeg"

}

"requestId":"6F338191-17FD-4A7C-ABA7-21D0AE87DADE"

}

百度识别的结果采用网络图片文字识别发现这个的识别度比通用文字识别度高

句对不起

讽刺了当初的爱你

返回的json字符串

{"direction": 0, "words_result_num": 2,"words_result": [ { "words": "句对不起", "probability": {"variance": 0.005448, "average": 0.95525, "min":0.827484 } }, { "words": "讽刺了当初的爱你", "probability": {"variance": 0.002724, "average": 0.953231, "min":0.868461 } } ] }

上面三个api虽然每个都没完整的识别出来不过目前看到还是百度的识别率最高不过阿里的有一个好处就是支持异步的因为图片转文字如果文字多的情况下转换时间还是比较长的对于实时性不高的项目来讲还是不错的不然需要自己实现异步比如用消息中间件让

单独一个系统处理转换然后在这个系统处理完再把结果返回

下面测试其他的图片