[{"data":1,"prerenderedAt":200},["ShallowReactive",2],{"guide:\u002Fguides\u002Fextract-text-from-image-local-ocr":3,"guides:list":117},{"id":4,"title":5,"body":6,"category":92,"description":93,"extension":94,"faq":95,"meta":105,"navigation":106,"path":107,"published":108,"seo":109,"stem":110,"toolSlugs":111,"updated":115,"__hash__":116},"guides\u002Fguides\u002Fextract-text-from-image-local-ocr.md","图片转文字：本地 OCR 的旋转、裁剪与校对方法",{"type":7,"value":8,"toc":84},"minimark",[9,18,22,35,38,41,54,57,60,63,66,70,73,81],[10,11,12,17],"p",{},[13,14,16],"a",{"href":15},"\u002Ftools\u002Fimage\u002Focr","图片文字识别","适合从清晰的中英文印刷文字截图、照片或扫描图片中提取纯文本。图片在本机浏览器内预处理和识别，不上传识别服务。它不还原原文版式，也不能把手写、表格或复杂排版的识别准确性作为保证，结果需要对照原图检查。",[19,20,21],"h2",{"id":21},"选择正确的输入",[10,23,24,25,29,30,34],{},"一次选择一张 PNG、JPG、WebP 或 BMP，最大 10 MiB、解码后不超过 1600 万像素。HEIC 请先在",[13,26,28],{"href":27},"\u002Ftools\u002Fimage\u002Fconvert?input=heic","图片格式转换的 HEIC 批量模式","中转成 JPG 或 PNG，下载后再手动添加到 OCR 页。需要处理扫描 PDF 时，先用 ",[13,31,33],{"href":32},"\u002Ftools\u002Fdocument\u002Fpdf-to-image","PDF 转图片","下载目标页，再手动选择到 OCR 页，不会自动转交文件。",[10,36,37],{},"文件超出像素限制时，应先用系统相册或尺寸工具缩小，再重新选择；OCR 页内的裁剪是在图片已被接受后使用，不能用它绕过输入上限。",[19,39,40],{"id":40},"先看实际识别预览",[42,43,44,48,51],"ol",{},[45,46,47],"li",{},"将文字旋转到正常阅读方向，可选顺时针 0、90、180 或 270 度。这里不是倾斜校正。",[45,49,50],{},"按上、右、下、左填写裁掉的百分比。各边允许 0–90%，上下合计和左右合计都须小于 100%。比例以旋转后的图片为准。",[45,52,53],{},"检查实际预览。工具先旋转和裁剪，再将最长边限制到 1600 px；这是实际送去识别的图像。",[10,55,56],{},"例如一张 4000×3000 的扫描图，整张识别会缩到约 1600×1200。若只需要其中一段小字，先把裁剪范围缩到那段文字，可减少无关内容；这不会补回模糊、反光或原图缺失的细节。",[19,58,59],{"id":59},"开始识别与首次下载",[10,61,62],{},"选择图片、旋转和裁剪都不会加载 OCR 引擎，点击「开始识别」后才下载。首次识别资源约 25 MB，加载和初始化速度取决于网络及设备；模型进度显示已解压的模型字节或缓存读取量，不等于网络压缩流量。初始化与识别阶段没有已知百分比时使用阶段提示。",[10,64,65],{},"公共模型可缓存在本机，但每次仍需初始化引擎，不能保证立即完成或长期离线可用。耗时较长可以取消；取消会终止 Worker，已有文字会保留。再次开始可重试，模型缓存不可用时仍会尝试网络加载。",[19,67,69],{"id":68},"校对编辑并保存","校对、编辑并保存",[10,71,72],{},"识别完成后重点检查数字、标点、英文大小写、人名和阅读顺序。结果可以直接编辑，「复制」和「下载 TXT」都使用当前编辑内容；TXT 不保留表格、字体或原始布局。",[10,74,75,76,80],{},"修改旋转或裁剪设置后，旧文字会保留并提示已过期。识别期间编辑文字时，新识别结果不会覆盖你的编辑；保存后需要时重新识别。",[77,78,79],"strong",{},"选择新图片会清空上一张结果","，请先下载。",[10,82,83],{},"图片和文字不自动保存。工具内「清理本站 OCR 模型缓存」只移除 OCR 专用 CacheStorage，下次识别会重新读取模型；浏览器 HTTP 缓存中的模型、脚本和运行时仍可能复用，必要时再联网下载。该操作不清理其他工具草稿。真实手机的内存和浏览器能力可能影响识别，遇到问题可缩小处理范围或换用较新的桌面浏览器。",{"title":85,"searchDepth":86,"depth":86,"links":87},"",2,[88,89,90,91],{"id":21,"depth":86,"text":21},{"id":40,"depth":86,"text":40},{"id":59,"depth":86,"text":59},{"id":68,"depth":86,"text":69},"image","从中英文截图或扫描图片提取纯文本，说明本地 OCR 的输入限制、实际预览、模型下载、文字编辑、取消、缓存和 TXT 保存步骤。","md",[96,99,102],{"q":97,"a":98},"能直接识别扫描 PDF 吗？","OCR 页只接受单张图片。可先用 PDF 转图片导出需要的页面，再逐张添加识别；PDF 处理工具本身不会自动 OCR。",{"q":100,"a":101},"能把表格还原为 Excel 吗？","不能。本站当前输出可编辑的纯文本，不保留表格结构、字体或复杂版面，不提供手写识别和 Excel 还原。",{"q":103,"a":104},"清理模型缓存会删除已编辑文字吗？","工具内的清理只删除 OCR 公共模型缓存，不会主动清空当前文字，也不清理其他草稿或浏览器 HTTP 缓存。但当前图片和文字不自动保存，离开或刷新前仍须复制或下载。",{},true,"\u002Fguides\u002Fextract-text-from-image-local-ocr",null,{"title":5,"description":93},"guides\u002Fextract-text-from-image-local-ocr",[112,113,114],"image\u002Focr","document\u002Fpdf-to-image","document\u002Ftext-count","2026-10-20","3EpvsxNONp085Ns9q4LiL9N5ioPvYDqwP56Q6bZugx8",[118,125,133,135,140,147,155,162,168,174,181,187,194],{"path":119,"title":120,"description":121,"category":92,"toolSlugs":122,"updated":124},"\u002Fguides\u002Fbatch-watermark-images","批量给图片加水印防搬运：文字水印与 Logo 水印实操","一次给几十张图片加文字或 Logo 水印：九宫格位置与平铺铺满怎么选、透明度旋转怎么调才既防搬运又不毁图，水印大小随图片宽度等比缩放的原理与批量导出步骤。",[123],"image\u002Fwatermark","2026-10-16",{"path":126,"title":127,"description":128,"category":92,"toolSlugs":129,"updated":132},"\u002Fguides\u002Fcompress-image-to-200kb","图片压到 200KB 以内的三种方法：按质量压、按目标体积压、转格式","上传证件、填报名表总卡在体积限制？三种思路把图片压到 200KB 以内：按质量逐档试、直接指定目标体积自动逼近、换 WebP\u002FJPG 格式再压，附批量操作与画质权衡建议。",[130,131],"image\u002Fcompress","image\u002Fconvert","2026-10-19",{"path":107,"title":5,"description":93,"category":92,"toolSlugs":134,"updated":115},[112,113,114],{"path":136,"title":137,"description":138,"category":92,"toolSlugs":139,"updated":115},"\u002Fguides\u002Fheic-to-jpg-local","HEIC 转 JPG 或 PNG：批量转换与下载步骤","在图片格式转换中选择 HEIC \u002F HEIF 批量模式，本地输出 JPG 或 PNG，核对预览并下载 ZIP。说明模式切换、文件限制、Live Photo 和失败重试边界。",[131,130],{"path":141,"title":142,"description":143,"category":144,"toolSlugs":145,"updated":132},"\u002Fguides\u002Fmake-qrcode-wifi-logo","二维码怎么做：网址码、WiFi 码、带中心 Logo 的定制码","三种最常见的二维码制作：网页链接码、连 WiFi 免输密码的 WiFi 码、带中心 Logo 的品牌码；配色与容错等级怎么配才既好看又扫得出，PNG\u002FSVG 导出与反向解析一次讲清。","developer",[146],"developer\u002Fqrcode",{"path":148,"title":149,"description":150,"category":151,"toolSlugs":152,"updated":124},"\u002Fguides\u002Fmd5-vs-sha256-file-checksum","MD5 和 SHA-256 怎么选：文件校验入门","下载文件怕被篡改、传完文件怕损坏？讲清 MD5 与 SHA-256 的差异与各自适用场景，文本与文件摘要怎么算、校验值怎么比对，以及分块流式处理大文件的实操步骤。","general",[153,154],"developer\u002Fmd5","security\u002Ffile-hash",{"path":156,"title":157,"description":158,"category":159,"toolSlugs":160,"updated":124},"\u002Fguides\u002Fmerge-split-pdf-scanned-docs","PDF 合并与拆分实操：把零散扫描件整理成一份干净文件","扫描件东一页西一页？手把手把多个 PDF 合并成一份、按页码范围拆分提取、删掉空白页与误扫页再导出，全程浏览器本地处理，合同与证件扫描件不出本机。","document",[161],"document\u002Fpdf-process",{"path":163,"title":164,"description":165,"category":159,"toolSlugs":166,"updated":124},"\u002Fguides\u002Fprint-copybook-tianzige","田字格字帖打印指南：从生字表到 A4 成品","在家给孩子打印练字帖：田字格\u002F米字格\u002F回宫格怎么选、拼音标注与描红模式怎么配、内容输什么（生字、姓名、古诗），A4 版式直接打印或另存 PDF 的完整流程。",[167],"document\u002Fcopybook",{"path":169,"title":170,"description":171,"category":92,"toolSlugs":172,"updated":124},"\u002Fguides\u002Fremove-exif-location-before-sharing","发照片前先清 EXIF：你的位置信息可能就藏在原图里","手机原图自带拍摄时间、机型、GPS 坐标等 EXIF 元数据。讲清哪些信息会泄露、哪些平台会清除哪些不会，以及发图前用浏览器本地工具一键查看并清除 EXIF 的完整步骤。",[173],"image\u002Fexif",{"path":175,"title":176,"description":177,"category":92,"toolSlugs":178,"updated":124},"\u002Fguides\u002Fresize-crop-image-for-platforms","一张图适配所有平台：改尺寸与裁剪比例速查","头图要 16:9、头像要 1:1？分清「改尺寸」与「裁剪」：等比缩放保持比例，按比例裁剪确定构图，附常见图片比例与批量处理流程，具体像素和格式以接收平台要求为准。",[179,180],"image\u002Fresize","image\u002Fcrop",{"path":182,"title":183,"description":184,"category":92,"toolSlugs":185,"updated":115},"\u002Fguides\u002Fstitch-images-to-long-image","多张截图拼成长图：顺序、统一尺寸与间距设置","用本地图片拼接工具横向或纵向合并截图，设置统一宽高、顺序、底色和输出格式。用具体尺寸示例解释长图大小，以及重复截图和超限的处理方式。",[186,180,130],"image\u002Fstitch",{"path":188,"title":189,"description":190,"category":191,"toolSlugs":192,"updated":124},"\u002Fguides\u002Ftrim-video-lossless-vs-reencode","视频怎么裁出一段：无损截取与重编码怎么选","只想留视频里 30 秒精华？对比无损快速与精确重编码裁剪，说明关键帧偏移、0.1 秒范围选择、引擎文件上限及输出编码的区别，附预览和下载步骤。","audio-video",[193],"audio-video\u002Fvideo-trim",{"path":195,"title":196,"description":197,"category":144,"toolSlugs":198,"updated":132},"\u002Fguides\u002Fwhat-is-unix-timestamp","Unix 时间戳是什么：秒与毫秒怎么区分、怎么换算","讲清 Unix 时间戳的秒与毫秒单位、自动识别的适用范围、时区差异和负数边界，附常见排查场景与手动指定单位的在线换算步骤。",[199],"developer\u002Ftimestamp",1792489886267]