Hey Arjun, let's fix that messy OCR output together! I see you're using Tesseract with Python but getting a lot of garbage characters—let's break down the issues and fix them step by step.

First, let's recap your current setup and the problematic output:

你的原代码 #

from PIL import Image
import pytesseract
image=Image.open('C://Users/Arjun/Desktop/1512350.jpg')
pytesseract.pytesseract.tesseract_cmd='C://Program Files (x86)/Tesseract- OCR/tesseract'
result=pytesseract.image_to_string(image,config='-psm7 -c tessedit_char_whitlist=ABCDEFGHIJKLMNOPQRSTUVWXYZ01234567890')
print(result)

识别输出（夹杂大量特殊字符） #

fl/'S‘TIW ° MILLER‘ 1003055666 ° gum/71; C6521:pmuzznmmimfmmpmy *5mg[e * 2900456023 ° Uj7s564550 ° 130013 ° mm 5\1£®IC/‘(L 0£0wEmm'2zowLI5vg gazmyw 250 0’/lrkksrmgf" flowzzyvg (jfiff-W" M * 42101 ° wowiany " qw— I’Va:/11/£172 ' J6 ’ 19955.65 * 5685.26 " 4586.65 ’ Safaxizf

1. 修正基础配置错误 #

You've got two small but critical typos in your config:

• tessedit_char_whitlist → tessedit_char_whitelist (missing an 'e' in whitelist)
• Your Tesseract path has an extra space: Tesseract- OCR → should be Tesseract-OCR (this might cause path resolution issues on Windows)

Fix those first, and that alone will help filter out some unwanted characters.

2. 图像预处理（最关键的优化步骤） #

Tesseract performs best on high-contrast, clean images. Let's add preprocessing steps to clean up your image:

• Convert to grayscale
• Apply binarization (convert to black/white to eliminate gray noise)
• Optional: Resize the image to increase resolution, or apply slight blur to reduce noise

Here's an updated code snippet with PIL-based preprocessing:

from PIL import Image, ImageOps
import pytesseract
import re

# 1. Load and preprocess the image
image = Image.open('C://Users/Arjun/Desktop/1512350.jpg')
# Convert to grayscale
gray_image = ImageOps.grayscale(image)
# Binarize (adjust threshold if needed—127 is default, tweak based on your image)
threshold = 127
binary_image = gray_image.point(lambda x: 0 if x < threshold else 255, '1')

# 2. Fix Tesseract config and path
pytesseract.pytesseract.tesseract_cmd = 'C://Program Files (x86)/Tesseract-OCR/tesseract'
# Use PSM 6 (assume a single uniform block of text) instead of 7 if your image has multiple lines
custom_config = r'-psm 6 -c tessedit_char_whitelist=ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789.'
result = pytesseract.image_to_string(binary_image, config=custom_config)

# 3. Post-process to remove any remaining unwanted characters
cleaned_result = re.sub(r'[^A-Z0-9.]', '', result)
print(cleaned_result)

If you have OpenCV installed, you can use more advanced preprocessing (like Gaussian blur to reduce noise):

import cv2
import numpy as np

image = cv2.imread('C://Users/Arjun/Desktop/1512350.jpg')
# Convert to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# Apply Gaussian blur to reduce noise
blurred = cv2.GaussianBlur(gray, (5,5), 0)
# Binarize with adaptive threshold (better for uneven lighting)
binary = cv2.adaptiveThreshold(blurred, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 11, 2)

# Convert back to PIL image for Tesseract
pil_image = Image.fromarray(binary)
# Rest of the Tesseract code same as above...

3. 调整PSM（页面分割模式） #

Your current -psm 7 tells Tesseract to treat the image as a single line of text. If your image has multiple lines or a block of text, switch to:

• -psm 6: Assume a single uniform block of text (great for forms, receipts, etc.)
• -psm 3: Default mode (automatically segments the page into text blocks)
• Tweak the PSM mode based on your image's layout—this can make a huge difference in accuracy.

4. 后处理保险措施 #

Even with perfect preprocessing, sometimes Tesseract slips through unwanted characters. Use Python's re module to strictly filter your output to only the characters you want:

import re
# Keep only uppercase letters, numbers, and periods (adjust regex based on your needs)
cleaned_result = re.sub(r'[^A-Z0-9.]', '', result)

额外提示 #

• Make sure you have the latest version of Tesseract installed (old versions have worse accuracy)
• If your text uses a rare font, you can train a custom Tesseract model, but that's more advanced—start with the steps above first.

内容的提问来源于stack exchange，提问作者Arjun Vc