静默截图OCR识别有误

[lajujuberry]

请先确认以下事项：

• 请务必查看 常见问题
• 已仔细阅读了 README
• 在 issues 页面搜索过问题（包括已关闭的 issue），但未能找到解决方法
• Easydict 已升级到 最新版本
• 我理解并认可上述内容，并理解项目维护者精力有限，不遵循规则的 issue 可能会被无视或直接关闭

问题描述

使用无法正常识别文本。
（只有“静默截图OCR”会乱码，“截图翻译”则正常。）
（此问题已出现数月，一直未恢复，特此反馈。）
（期间我有试过重启电脑等其他一系列常规的解决操作，无变化。）

该问题是否可以稳定重现？

可重现

重现步骤

1. 通过快捷键启用“静默截图OCR”
2. 框选
3. 粘贴后可见文本乱码

截图

比如：启用静默截图OCR，我框选下图文字部分（框选时有避开图标，只框了“DeepL 翻译”这一部分）

结果为：DeepL #1 i7

运行日志（zip 文件）

Easydict log 2025-12-07 上午11-19-55-615.zip

期望结果

OCR正常。

解决方案（可选）

No response

设备型号 && 系统版本

Apple M4 Pro, Tahoe 26.1

是否愿意提交 PR 解决该问题？

• 我愿意提交 PR 来解决该问题

[github-actions]

Hello lajujuberry, Thank you for your first issue contribution 🎉

[uuuhvfun]

+1 之前的静默ocr特别好用 还可以获取二维码链接，中英扫描0错

[tisfeng]

（期间我有试过重启电脑等其他一系列常规的解决操作，无变化。）

这个问题很奇怪，看起来看 #1019 有点类似，但你重启电脑却没有解决，很迷惑。

[tisfeng]

这张图片，在我电脑上测试，直接剪贴板翻译，或者静默 OCR，都能正确识别：DeepL 翻译

ocr result: DeepL 翻译

[2025-12-21 11:49:04.079 ● EZWindowManager ● 896 ● ℹ️] -[EZWindowManager silentScreenshotOCR] ● Silent screenshot and OCR
[2025-12-21 11:49:04.079 ● EZWindowManager ● 950 ● ℹ️] -[EZWindowManager captureWithRestorePreviousApp:completion:] ● Starting capture
[2025-12-21 11:49:06.860 ● EZWindowManager ● 971 ● ℹ️] -[EZWindowManager captureWithRestorePreviousApp:completion:]_block_invoke ● Screenshot captured: <NSImage 0x9b54694a0 Size={164, 56} RepProvider=<NSImageArrayRepProvider: 0x9b06c6e20, reps:(
    "<NSCGImageSnapshotRep:0x9b1472300 cgImage=<CGImage 0x9b505d2c0> (DP)\n\t<<CGColorSpace 0x9b28d4300> (kCGColorSpaceICCBased; kCGColorSpaceModelRGB; \U5f69\U8272LCD)>\n\t\theadroom = 1.000000  \n\t\twidth = 164, height = 56, bpc = 8, bpp = 32, row bytes = 13824, \n\t\tkCGImageAlphaPremultipliedFirst | kCGImageByteOrder32Little  | kCGImagePixelFormatPacked \n\t\tis mask? No, has masking color? No, has soft mask? No, has matte? No, should interpolate? No>"
)>>
[2025-12-21 11:49:06.862 ● EZWindowManager ● 981 ● ℹ️] -[EZWindowManager captureWithRestorePreviousApp:completion:]_block_invoke ● Saved image: /Users/tisfeng/Library/Caches/com.izual.Easydict/MMLogs/Image/snip_image.png
[2025-12-21 11:49:06.862 ● EZBaseQueryViewController ● 508 ● ℹ️] -[EZBaseQueryViewController startOCRImage:actionType:autoQuery:] ● start OCR Image: NSSize: {164, 56}, actionType: silent_screenshot_ocr
[2025-12-21 11:49:06.862 ● AppleOCREngine ● 48 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Recognizing text in image with language: 简体中文(zh-Hans), image size: (164.0, 56.0)
[2025-12-21 11:49:06.863 ● AppleOCREngine ● 424 ● ℹ️] performVisionOCR(on:language:) ● Using modern RecognizeTextRequest API
[2025-12-21 11:49:06.863 ● AppleOCREngine ● 461 ● ℹ️] performSingleModernVisionOCR(on:language:) ● Performing modern Vision OCR with language: 简体中文(zh-Hans), auto detect: false
[2025-12-21 11:49:07.205 ● AppleOCREngine ● 68 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Recognize observations count: 1 (简体中文(zh-Hans))
[2025-12-21 11:49:07.205 ● AppleOCREngine ● 69 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Cost time: 0.342 seconds
[2025-12-21 11:49:07.210 ● ChineseGenreAnalyzer ● 35 ● ℹ️] detect() ● 
Starting Chinese text detection...
[2025-12-21 11:49:07.210 ● String+Utils ● 42 ● ℹ️] calculateLinguisticFeatureRatio(for:) ● Found linguistic features: []
[2025-12-21 11:49:07.210 ● String+Utils ● 42 ● ℹ️] calculateLinguisticFeatureRatio(for:) ● Found linguistic features: []
[2025-12-21 11:49:07.210 ● ChineseGenreAnalyzer ● 51 ● ℹ️] detect() ● Text count (2) is less than 20, or phrase count (1) is less than 4, skipping classical Chinese detection.
[2025-12-21 11:49:07.210 ● AppleOCREngine ● 78 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Detected language: 简体中文(zh-Hans), probabilities: 
zh-Hans: 1.000
ja: 0.000
zh-Hant: 0.000
[2025-12-21 11:49:07.210 ● AppleOCREngine ● 92 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Merged text char count: 8
[2025-12-21 11:49:07.210 ● AppleOCREngine ● 93 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Performing OCR text processing, smart merging: true
[2025-12-21 11:49:07.210 ● AppleOCREngine ● 114 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● OCR completion (简体中文(zh-Hans)) cost time: 0.347 seconds
[2025-12-21 11:49:07.210 ● EZBaseQueryViewController ● 524 ● ℹ️] -[EZBaseQueryViewController startOCRImage:actionType:autoQuery:]_block_invoke ● ocr result: DeepL 翻译

[tisfeng]

等等，看了你提供的日志，你这里 OCR 为什么使用的是英语 en ？是你手动将查询语言修改为了英语吗？

建议不要手动修改这里的查询语言，让程序自己识别就好，除非你知道你在做什么。

DeepL 翻译

这个让它自动识别为中文就好，如果你手动修改过，请重新设置为自动识别试试。

这里也可能有 bug，稍后我检查一下代码。

[2025-12-07 上午11:15:52.413 ● EZBaseQueryViewController ● 508 ● ℹ️] -[EZBaseQueryViewController startOCRImage:actionType:autoQuery:] ● start OCR Image: NSSize: {166, 70}, actionType: silent_screenshot_ocr
[2025-12-07 上午11:15:52.414 ● AppleOCREngine ● 48 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Recognizing text in image with language: 英语(en), image size: (166.0, 70.0)
[2025-12-07 上午11:15:52.415 ● AppleOCREngine ● 424 ● ℹ️] performVisionOCR(on:language:) ● Using modern RecognizeTextRequest API
[2025-12-07 上午11:15:52.415 ● AppleOCREngine ● 461 ● ℹ️] performSingleModernVisionOCR(on:language:) ● Performing modern Vision OCR with language: 英语(en), auto detect: false
[2025-12-07 上午11:15:52.652 ● AppleOCREngine ● 68 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Recognize observations count: 1 (英语(en))
[2025-12-07 上午11:15:52.652 ● AppleOCREngine ● 69 ● ℹ️] recognizeText(image:language:requiresAccurateRecognition:) ● Cost time: 0.237 seconds

[lajujuberry]

确定不是语言选择的问题。
目前是“自动检测”，再次尝试静默截取例图，
ocr result第一次: DeepL #i·
ocr result第二次: DeepL #i7
它会变。

失灵的情况越来越多，
静默OCR常常识别不出任何东西。
划词翻译也划不上词（选中文本后，通过划词翻译快捷键只会打开easydict窗口，内容是空的）。