dsh-image-ocr 0.0.0-stage → 0.0.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +39 -2
- package/package.json +28 -4
package/README.md
CHANGED
|
@@ -1,3 +1,40 @@
|
|
|
1
|
-
#
|
|
1
|
+
# dsh-image-ocr · 图片与扫描件 OCR
|
|
2
2
|
|
|
3
|
-
|
|
3
|
+
> **定义**:把图片、截图、扫描版 PDF 里的文字、表格结构和坐标置信度抽出来,并带上出处(第几页/哪一块/什么引擎),让 agent 能引用而不是猜。
|
|
4
|
+
>
|
|
5
|
+
> **English**: Extract text, table structure and per-box confidence from images, screenshots and scanned PDFs, with provenance (page, region, engine) so an agent can cite instead of guess.
|
|
6
|
+
|
|
7
|
+
状态:**0.0.1 名称占位**(name reservation)。实现中;本文件写的是这个包**要做的事与边界**,不是已完成的能力。
|
|
8
|
+
|
|
9
|
+
## 它解决什么
|
|
10
|
+
|
|
11
|
+
把图片、截图、扫描版 PDF 里的文字、表格结构和坐标置信度抽出来,并带上出处(第几页/哪一块/什么引擎),让 agent 能引用而不是猜。
|
|
12
|
+
|
|
13
|
+
本项目已有 dsh-doc-parse 负责 PDF/Word/Excel 的**结构化解析**,但没有 OCR:扫描件与截图里的字它读不到。这个包补的正是那一段。
|
|
14
|
+
|
|
15
|
+
## 计划能力(0.1.0 目标)
|
|
16
|
+
|
|
17
|
+
- 图片/截图/扫描 PDF → 文本(中英混排)
|
|
18
|
+
- 表格结构(行列合并)与每个文本块的 bbox + 置信度
|
|
19
|
+
- 低置信度逐块列出,不静默丢字
|
|
20
|
+
- 多引擎可切换(本地 tesseract / 云端 API),默认不联网
|
|
21
|
+
|
|
22
|
+
## 边界(明确不做)
|
|
23
|
+
|
|
24
|
+
- 不把「占名」说成「有人在用」:本包当前只是一个名字。
|
|
25
|
+
- 不替代专业判断:输出是**可核对的清单**,不是最终结论。
|
|
26
|
+
- 不静默编造:取不到的数据标 `unknown` 并说明原因。
|
|
27
|
+
|
|
28
|
+
## 安装(0.1.0 发布后)
|
|
29
|
+
|
|
30
|
+
```sh
|
|
31
|
+
dsh plugin --profile desktop add dsh-image-ocr
|
|
32
|
+
```
|
|
33
|
+
|
|
34
|
+
## 相关包
|
|
35
|
+
|
|
36
|
+
本包属于 [dsh-namepool](https://github.com/dongsheng123132/dsh-namepool) 插件族(47+ 个零依赖 DSH 插件)。
|
|
37
|
+
|
|
38
|
+
## 许可
|
|
39
|
+
|
|
40
|
+
MIT
|
package/package.json
CHANGED
|
@@ -1,6 +1,30 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "dsh-image-ocr",
|
|
3
|
-
"version": "0.0.
|
|
4
|
-
"
|
|
5
|
-
"
|
|
6
|
-
|
|
3
|
+
"version": "0.0.1",
|
|
4
|
+
"description": "Extract text, table structure and per-box confidence from images, screenshots and scanned PDFs, with provenance (page, region, engine) so an agent can cite instead of guess. | 把图片、截图、扫描版 PDF 里的文字、表格结构和坐标置信度抽出来,并带上出处(第几页/哪一块/什么引擎),让 agent 能引用而不是猜。",
|
|
5
|
+
"keywords": [
|
|
6
|
+
"dsh",
|
|
7
|
+
"dsh-plugin",
|
|
8
|
+
"deepseek-harness",
|
|
9
|
+
"agent",
|
|
10
|
+
"ocr",
|
|
11
|
+
"image-text",
|
|
12
|
+
"scanned-pdf",
|
|
13
|
+
"table-extraction",
|
|
14
|
+
"bounding-box",
|
|
15
|
+
"confidence",
|
|
16
|
+
"screenshot",
|
|
17
|
+
"invoice",
|
|
18
|
+
"text-recognition",
|
|
19
|
+
"provenance",
|
|
20
|
+
"扫描件"
|
|
21
|
+
],
|
|
22
|
+
"license": "MIT",
|
|
23
|
+
"type": "module",
|
|
24
|
+
"files": [
|
|
25
|
+
"README.md"
|
|
26
|
+
],
|
|
27
|
+
"engines": {
|
|
28
|
+
"node": ">=22"
|
|
29
|
+
}
|
|
30
|
+
}
|