OCR Server

Turn your iPhone into a powerful local OCR server using Apple's Vision Framework. No cloud dependencies, unlimited usage, complete privacy.

Download from the App Store

English | 日本語 | 繁體中文 | 简体中文 | 한국어 | Français

How to Use

1. Launch the app and the server will start automatically
2. Access the displayed IP address from any device on the same network
3. Upload images to get text recognition results
4. Integrate the service into your applications via API
5. To ensure the app runs continuously without interruption, please enable iOS Guided Access mode and keep the screen on

• OCR Test: On your computer, open a web browser and navigate to the IP address displayed by the app to perform an OCR test.

• API Example - Upload an image via upload API:

  curl -H "Accept: application/json" \
    -X POST http://<YOUR IP>:8000/upload \
    -F "file=@01.png"

• Python Upload Example:

  import requests
  
  url = "http://10.0.1.11:8000/upload"  # Replace with your IP address
  file_path = "01.png"
  
  with open(file_path, "rb") as f:
      files = {"file": f}
      headers = {"Accept": "application/json"}
      response = requests.post(url, files=files, headers=headers)
  
  print("status code:", response.status_code)
  print("response:", response.text)

• The JSON response looks like this:

  {
    "success": true,
    "message": "File uploaded successfully",
    "ocr_result": "Hello\nWorld",
    "image_height": 648,
    "image_width": 1247,
    "ocr_boxes": [
      {
        "text": "Hello",
        "x": 429.6554479416482,
        "y": 268.0000001076923,
        "w": 201.83814102564105,
        "h": 72,
        "rect": {
          "topLeft_x": 429.6554479416482,
          "topLeft_y": 268.0000001076923,
          "topRight_x": 631.4935889672893,
          "topRight_y": 268.0000001076923,
          "bottomRight_x": 631.4935889672893,
          "bottomRight_y": 340.0000001076923,
          "bottomLeft_x": 429.6554479416482,
          "bottomLeft_y": 340.0000001076923
        }
      },
      {
        "text": "World",
        "x": 421.6618595738782,
        "y": 417.99999971428576,
        "w": 251.79807692307696,
        "h": 80,
        "rect": {
          "topLeft_x": 421.6618595738782,
          "topLeft_y": 417.99999971428576,
          "topRight_x": 673.4599364969552,
          "topRight_y": 417.99999971428576,
          "bottomRight_x": 673.4599364969552,
          "bottomRight_y": 497.99999971428576,
          "bottomLeft_x": 421.6618595738782,
          "bottomLeft_y": 497.99999971428576
        }
      }
    ]
  }

  image_width and image_height represent the width and height of the image (in px), x and y represent the top-left origin of the text bounding box (in px), w and h represent the width and height of the text bounding box (in px), rect provides the four corner coordinates of the detected text region, preserving its original orientation (non-axis-aligned).

• Python Example – Drawing text bounding boxes using ocr_boxes information:

  #
  # pip3 install requests pillow opencv-python numpy
  #
  
  import os
  import sys
  import requests
  from PIL import Image, ImageDraw, ImageFont, ImageOps
  import numpy as np
  import cv2
  
  url = "http://10.0.1.11:8000/upload"  # Replace with your IP address
  file_path = "01.png"
  
  # ===== Select font (supports Chinese and English), font size auto-scales with box height =====
  def pick_font(box_h_px: float):
      font_candidates = [
          # macOS
          "/System/Library/Fonts/PingFang.ttc",
          "/System/Library/Fonts/STHeiti Light.ttc",
          # Windows
          r"C:\Windows\Fonts\msyh.ttc",
          r"C:\Windows\Fonts\msjh.ttc",
          r"C:\Windows\Fonts\arialuni.ttf",
          # Noto
          "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
          "/usr/share/fonts/truetype/noto/NotoSansCJK-Regular.ttc",
      ]
      size = max(10, int(box_h_px * 0.25))  # Small font size = 25% of box height (minimum 10pt)
      for path in font_candidates:
          if os.path.exists(path):
              try:
                  return ImageFont.truetype(path, size=size)
              except Exception:
                  pass
      return ImageFont.load_default()
  
  # ===== Draw box and small text =====
  def draw_boxes(img_pil: Image.Image, boxes, line_thickness: int = 5) -> Image.Image:
      draw = ImageDraw.Draw(img_pil)
      for b in boxes:
          try:
              x = float(b["x"]); y = float(b["y"])
              w = float(b["w"]); h = float(b["h"])
              text = str(b.get("text", ""))
          except Exception:
              continue
  
          # Red bounding box
          x2, y2 = x + w, y + h
          draw.rectangle([x, y, x2, y2], outline=(255, 0, 0), width=line_thickness)
  
          # Top-right label
          font = pick_font(h)
          # Text size
          # textbbox returns (l, t, r, b)
          l, t, r, b = draw.textbbox((0, 0), text, font=font)
          tw, th = (r - l), (b - t)
          pad = max(2, int(h * 0.06))
  
          # Align label to top-right, not exceeding box or image edge
          tx = int(max(0, min(x2 - tw - pad, img_pil.width - tw - pad)))
          ty = int(max(0, min(y + pad, img_pil.height - th - pad)))
  
          # White background
          draw.rectangle([tx - pad, ty - pad, tx + tw + pad, ty + th + pad], fill=(255, 255, 255))
          draw.text((tx, ty), text, font=font, fill=(20, 20, 20))
      return img_pil
  
  def main():
      if not os.path.exists(file_path):
          print(f"[ERROR] Image not found: {file_path}", file=sys.stderr)
          sys.exit(1)
  
      # 1) Upload
      with open(file_path, "rb") as f:
          files = {"file": f}
          headers = {"Accept": "application/json"}
          try:
              response = requests.post(url, files=files, headers=headers, timeout=60)
          except requests.RequestException as e:
              print(f"[ERROR] Request failed: {e}", file=sys.stderr)
              sys.exit(2)
  
      print("status code:", response.status_code)
  
      # 2) Check HTTP and JSON
      if response.status_code != 200:
          print("response:", response.text[:500])
          sys.exit(3)
  
      try:
          data = response.json()
      except ValueError:
          print("[ERROR] Not JSON response")
          print("response:", response.text[:500])
          sys.exit(4)
  
      if not data.get("success", False):
          print("[ERROR] Server returned failure:", data)
          sys.exit(5)
  
      print("response ok")
  
      # 3) Load original image (using PIL)
      img_pil = Image.open(file_path)
      img_pil = ImageOps.exif_transpose(img_pil).convert("RGB")
  
      # If server returns different dimensions (should usually match), use server dimensions
      W = int(data.get("image_width", img_pil.width))
      H = int(data.get("image_height", img_pil.height))
      if (W, H) != (img_pil.width, img_pil.height):
          img_pil = img_pil.resize((W, H), Image.BICUBIC)
  
      boxes = data.get("ocr_boxes", [])
      img_pil = draw_boxes(img_pil, boxes)
  
      # 4) Display
      img_cv = cv2.cvtColor(np.array(img_pil), cv2.COLOR_RGB2BGR)
      cv2.imshow("OCR Preview", img_cv)
      print("Press any key on the image window to exit...")
      cv2.waitKey(0)
      cv2.destroyAllWindows()
  
  if __name__ == "__main__":
      main()

  Sample Output:

  

Document Paragraph Detection (iOS 26+ Only)

On the webpage, check Document Paragraph Detection:

Paragraph recognition results:

Or recognize document paragraphs using the docOCR API:

curl -H "Accept: application/json" \
    -X POST http://<YOUR IP>:8000/docOCR \
    -F "file=@01.png"

The JSON response looks like this:

{
    "success": true,
    "message": "OCR completed successfully",
    "ocr_text": "OCR text..."
}

Features

• High-precision OCR powered by Apple’s Vision Framework
• Supports multiple languages with automatic detection
• Upload via web interface and receive OCR results within seconds
• JSON API for easy integration into apps
• 100% local processing, no cloud, full privacy

Use Cases

• Local OCR without cloud services
• Share OCR services across devices in the same network
• Build an OCR processing cluster using multiple iPhones