Improving OCR accuracy

Question

I am trying to write a code to read a game's log and send a message to discord based on its content. So far the code works fine but I am having trouble with the OCR. Sometimes its bad accuracy causes messages to be sent more than 1 time. I need the OCR to be as exact as possible to avoid duplication of messages. I researched and find out to do a binarization/thresholding process to get black text on a white background before using trying OCR although this just worsen the results. What strategies can I use to improve its accuracy in this case? The log looks like this:

The code is the following:

import cv2
import numpy as np
import pytesseract
import time
import requests
import json
import pyautogui

# Discord webhook URL
webhook_url = "https://discord.com/api/webhooks/1111838770568888401/GJeNJ9YrMgq6_Mn3XMcX2JQlvmlZsmGhNLp1sNSRpu1RxRXEwkctIUe_iRanurRz1pS8"

# Phrases to monitor and their corresponding Discord messages
target_phrases = {
    "was killed": "<@&1184910793871994920> Day{message}",
    "Tamed": "<@&1184962484675809340> Day{message}",
}

# OCR configuration
tesseract_path = r"C:\Program Files\Tesseract-OCR\tesseract.exe"
pytesseract.pytesseract.tesseract_cmd = tesseract_path

# Time interval between OCR checks (in seconds)
check_interval = 5

# Set to store previous occurrences of the target phrases
previous_occurrences = set()

# Define the screen region where the list is located
list_region = (750, 198, 400, 400)  # Replace with the coordinates of the list region

# Function to send a message to Discord webhook
def send_discord_message(message):
    data = {"content": message}
    headers = {"Content-Type": "application/json"}
    response = requests.post(webhook_url, data=json.dumps(data), headers=headers)
    if response.status_code != 204:
        print("Failed to send Discord message:", response.text)
    else:
        print("Discord message sent successfully")

# Main loop
while True:
    # Take a screenshot of the defined screen region
    screenshot = pyautogui.screenshot(region=list_region)
    screenshot = cv2.cvtColor(np.array(screenshot), cv2.COLOR_RGB2BGR)

    # Apply OCR to extract text from the screenshot
    extracted_text = pytesseract.image_to_string(screenshot)
    print("Extracted Text:\n", extracted_text)

    # Split the extracted text into messages based on the occurrence of "Day"
    messages = extracted_text.strip().split('Day')

    # Process each message individually
    for message in messages:
        # Check if any target phrase appears in the message
        for target_phrase, discord_message in target_phrases.items():
            if target_phrase.lower() in message.lower() and message.strip() not in previous_occurrences:
                # Send a message for the new occurrence
                print(f"Phrase '{target_phrase}' detected")
                send_discord_message(discord_message.format(message=message))

                # Update the set of previous occurrences
                previous_occurrences.add(message.strip())

    # Wait for the next OCR check
    time.sleep(check_interval)

Answer 1

We can improving the accuracy of the ocr output result by below two ways.

1. Using tessdata-best or tessdata-fast.
2. Improving the input image quality by third party library. Most of when we increasing the resolution of image, the better will be returned.

Please find the tessdata link below,

• https://github.com/tesseract-ocr/tessdata_fast
• https://github.com/tesseract-ocr/tessdata_best

refer the tessdata path by using the --tessdata-dir argument