DEV Community

jack zhu
jack zhu

Posted on

Is Python OCR Inaccurate? Try These Image Preprocessing Techniques!

Is Python OCR Inaccurate? Try These Image Preprocessing Techniques!

When using Python for OCR (Optical Character Recognition), poor image quality — such as blur, skew, or noise — can lead to low recognition accuracy. This article introduces essential image preprocessing techniques to improve OCR performance, along with recommended third-party image enhancement APIs.

✅ 1. Key Image Preprocessing Techniques

1. Adaptive Thresholding to Enhance Contrast

Use adaptive thresholding to handle uneven lighting or background:

import cv2


img = cv2.imread('input.jpg', 0)

binary = cv2.adaptiveThreshold(

    img, 255,

    cv2.ADAPTIVE_THRESH_GAUSSIAN_C,

    cv2.THRESH_BINARY, 11, 2

)

cv2.imwrite('binary.jpg', binary)
Enter fullscreen mode Exit fullscreen mode



  1. Denoising and Removing Artifacts 

blur = cv2.GaussianBlur(binary, (3, 3), 0)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (2, 2))
denoised = cv2.morphologyEx(blur, cv2.MORPH_OPEN, kernel)
cv2.imwrite('denoised.jpg', denoised)
Enter fullscreen mode Exit fullscreen mode

  • Deskewing: Correct Image Rotation 
    import numpy as np



    coords = cv2.findNonZero(denoised)
    
angle = cv2.minAreaRect(coords)[-1]
    


    if angle < -45:
    
    angle = -(90 + angle)
    
else:
    
    angle = -angle
    


    (h, w) = denoised.shape[:2]
    
center = (w // 2, h // 2)
    
M = cv2.getRotationMatrix2D(center, angle, 1.0)
    
rotated = cv2.warpAffine(denoised, M, (w, h), flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_REPLICATE)
    


    cv2.imwrite('rotated.jpg', rotated)
    Enter fullscreen mode Exit fullscreen mode



    1. Upscaling Low-Resolution Images

    • Bicubic interpolation is recommended to retain text clarity:

    resized = cv2.resize(rotated, None, fx=2, fy=2, interpolation=cv2.INTER_CUBIC)
    
cv2.imwrite('resized.jpg', resized)
    Enter fullscreen mode Exit fullscreen mode



    1. Shadow and Uneven Lighting Removal 

    dilated = cv2.dilate(img, np.ones((7,7), np.uint8))
bg = cv2.medianBlur(dilated, 21)
diff = 255 - cv2.absdiff(img, bg)
norm = cv2.normalize(diff, None, alpha=0, beta=255, norm_type=cv2.NORM_MINMAX)
cv2.imwrite('shadow_removed.jpg', norm)
    Enter fullscreen mode Exit fullscreen mode

    🚀 2. Recommended Image Enhancement APIs

    To simplify preprocessing, consider using these high-quality online tools and APIs:

    📐 Document Correction

    Auto deskew and perspective correction

    📄 Virtual Scanner

    Scan-like enhancement & background removal

    🌫️ Shadow Removal

    Fix uneven lighting and shadows

    🔍 Image Enhancement

    Improve sharpness, contrast, brightness

    🌐 OCR + Translation

    Extract and translate text automatically

    Visit API

    🧠 3. Recommended OCR Processing Workflow

    1. 📤 Upload original image → Process using API (deskew, denoise, etc.)
    2. 📥 Download enhanced image
    3. ⚙️ Apply further preprocessing if needed (thresholding, resize, etc.)
    4. 🔠 Use OCR engine (Tesseract / PaddleOCR) to extract text
    5. 🧹 Post-process output (correct errors, restore layout)

    Need a full Python template to automate preprocessing and OCR? Let me know — I can provide a complete script.

    Top comments (0)