注
PDF、Office ファイル、HTML ドキュメント、ドキュメント イメージからテキストを抽出するには、 ドキュメント インテリジェンス読み取り OCR モデルを使用します。 それは、テキストが多いデジタル ドキュメントとスキャンされたドキュメント用に最適化されており、それを容易にしてインテリジェント ドキュメント処理のシナリオを支援する非同期 API が使われています。
OCR は、製品のラベル、ユーザーが生成した画像、スクリーンショット、道路標識、ポスターなど、整えられていない、ドキュメント以外の画像からテキストを抽出するための、機械学習ベースの手法です。 Azure AI Vision の OCR サービスには、画像に含まれるテキストが多くない軽量のシナリオのための、高速の同期 API があります。 このアプローチを使用すると、ほぼリアルタイムのユーザー エクスペリエンスに OCR を埋め込んで、コンテンツの理解を深め、ユーザー アクションを迅速なターンアラウンド時間でフォローアップすることができます。
Azure AI Vision v4.0 Read OCR とは
新しい Azure AI Vision Image Analysis 4.0 REST API を使うと、パフォーマンスが向上した統合同期 API で画像から印刷または手書きのテキストを抽出でき、1 回の API 操作で、OCR の結果を含むすべての画像分析情報を簡単に取得できます。 Read OCR エンジンは、グローバル言語サポート用のユニバーサル スクリプト ベースのモデルでサポートされる複数のディープ ラーニング モデルに基づいて構築されています。
テキスト抽出の例
次の JSON 応答は、指定した画像からテキストを抽出するときに、Image Analysis 4.0 API が返す内容を示しています。
{
"modelVersion": "2024-02-01",
"metadata":
{
"width": 1000,
"height": 945
},
"readResult":
{
"blocks":
[
{
"lines":
[
{
"text": "You must be the change you",
"boundingPolygon":
[
{"x":251,"y":265},
{"x":673,"y":260},
{"x":674,"y":308},
{"x":252,"y":318}
],
"words":
[
{"text":"You","boundingPolygon":[{"x":252,"y":267},{"x":307,"y":265},{"x":307,"y":318},{"x":253,"y":318}],"confidence":0.996},
{"text":"must","boundingPolygon":[{"x":318,"y":264},{"x":386,"y":263},{"x":387,"y":316},{"x":319,"y":318}],"confidence":0.99},
{"text":"be","boundingPolygon":[{"x":396,"y":262},{"x":432,"y":262},{"x":432,"y":315},{"x":396,"y":316}],"confidence":0.891},
{"text":"the","boundingPolygon":[{"x":441,"y":262},{"x":503,"y":261},{"x":503,"y":312},{"x":442,"y":314}],"confidence":0.994},
{"text":"change","boundingPolygon":[{"x":513,"y":261},{"x":613,"y":262},{"x":613,"y":306},{"x":513,"y":311}],"confidence":0.99},
{"text":"you","boundingPolygon":[{"x":623,"y":262},{"x":673,"y":263},{"x":673,"y":302},{"x":622,"y":305}],"confidence":0.994}
]
},
{
"text": "wish to see in the world !",
"boundingPolygon":
[
{"x":325,"y":338},
{"x":695,"y":328},
{"x":696,"y":370},
{"x":325,"y":381}
],
"words":
[
{"text":"wish","boundingPolygon":[{"x":325,"y":339},{"x":390,"y":337},{"x":391,"y":380},{"x":326,"y":381}],"confidence":0.992},
{"text":"to","boundingPolygon":[{"x":406,"y":337},{"x":443,"y":335},{"x":443,"y":379},{"x":407,"y":380}],"confidence":0.995},
{"text":"see","boundingPolygon":[{"x":451,"y":335},{"x":494,"y":334},{"x":494,"y":377},{"x":452,"y":379}],"confidence":0.996},
{"text":"in","boundingPolygon":[{"x":502,"y":333},{"x":533,"y":332},{"x":534,"y":376},{"x":503,"y":377}],"confidence":0.996},
{"text":"the","boundingPolygon":[{"x":542,"y":332},{"x":590,"y":331},{"x":590,"y":375},{"x":542,"y":376}],"confidence":0.995},
{"text":"world","boundingPolygon":[{"x":599,"y":331},{"x":664,"y":329},{"x":664,"y":372},{"x":599,"y":374}],"confidence":0.995},
{"text":"!","boundingPolygon":[{"x":672,"y":329},{"x":694,"y":328},{"x":694,"y":371},{"x":672,"y":372}],"confidence":0.957}
]
},
{
"text": "Everything has its beauty , but",
"boundingPolygon":
[
{"x":254,"y":439},
{"x":644,"y":433},
{"x":645,"y":484},
{"x":255,"y":488}
],
"words":
[
{"text":"Everything","boundingPolygon":[{"x":254,"y":442},{"x":379,"y":440},{"x":380,"y":486},{"x":257,"y":488}],"confidence":0.97},
{"text":"has","boundingPolygon":[{"x":388,"y":440},{"x":435,"y":438},{"x":436,"y":485},{"x":389,"y":486}],"confidence":0.965},
{"text":"its","boundingPolygon":[{"x":445,"y":438},{"x":485,"y":437},{"x":486,"y":485},{"x":446,"y":485}],"confidence":0.99},
{"text":"beauty","boundingPolygon":[{"x":495,"y":437},{"x":567,"y":435},{"x":568,"y":485},{"x":496,"y":485}],"confidence":0.685},
{"text":",","boundingPolygon":[{"x":577,"y":435},{"x":583,"y":435},{"x":583,"y":485},{"x":577,"y":485}],"confidence":0.939},
{"text":"but","boundingPolygon":[{"x":589,"y":435},{"x":644,"y":434},{"x":644,"y":485},{"x":589,"y":485}],"confidence":0.628}
]
},
{
"text": "not everyone sees it !",
"boundingPolygon":
[
{"x":363,"y":508},
{"x":658,"y":493},
{"x":659,"y":539},
{"x":364,"y":552}
],
"words":
[
{"text":"not","boundingPolygon":[{"x":363,"y":510},{"x":412,"y":508},{"x":413,"y":548},{"x":365,"y":552}],"confidence":0.989},
{"text":"everyone","boundingPolygon":[{"x":420,"y":507},{"x":521,"y":501},{"x":522,"y":542},{"x":421,"y":548}],"confidence":0.924},
{"text":"sees","boundingPolygon":[{"x":536,"y":501},{"x":588,"y":498},{"x":589,"y":540},{"x":537,"y":542}],"confidence":0.987},
{"text":"it","boundingPolygon":[{"x":597,"y":497},{"x":627,"y":495},{"x":628,"y":540},{"x":598,"y":540}],"confidence":0.995},
{"text":"!","boundingPolygon":[{"x":635,"y":495},{"x":656,"y":494},{"x":657,"y":540},{"x":636,"y":540}],"confidence":0.952}
]
}
]
}
]
}
}
API の使用
テキスト抽出機能は、Analyze Image API の一部です。
Read を features クエリ パラメーターに追加します。 次に、完全な JSON 応答が得られたら、"readResult" セクションのコンテンツの文字列を解析します。
次の手順
Image Analysis 4.0 API を使用して画像からテキストを抽出するには、「画像分析のクイックスタート」に従ってください。