How can I implement a basic digit recognition OCR system in OpenCV-Python using KNearest and SVM algorithms?

Simple Digit Recognition OCR in OpenCV-Python

Introduction

This article aims to guide you through the implementation of a basic digit recognition OCR (Optical Character Recognition) system using OpenCV-Python. We'll explore two popular machine learning algorithms: KNearest and SVM.

Question 1: Letter_recognition.data File

Letter_recognition.data is a dataset included in OpenCV-Python samples. It contains a collection of handwritten letters along with 16 feature values for each letter. This file serves as training data for various character recognition tasks.

Building Your Own Letter_recognition.data:

You can create your own letter_recognition.data file by following these steps:

1. Prepare your letter dataset with each letter represented as a 10x10 pixel image.
2. Extract pixel values from each image to form a feature vector of 100 values.
3. Manually assign a label (0-25, corresponding to A-Z) to each letter.
4. Save the feature vectors and labels in a text file, with each row in the format: ,,,...,

Question 2: results.ravel() in KNearest

results.ravel() converts the array of recognized digits from a multi-dimensional array to a flat 1D array. This makes it easier to interpret and display the results.

Question 3: Simple Digit Recognition Tool

To create a simple digit recognition tool using letter_recognition.data, follow these steps:

Data Preparation:

• Load your custom letter_recognition.data file or use the sample from OpenCV.

Training:

• Create a KNearest or SVM classifier instance.
• Train the classifier using the samples and responses from letter_recognition.data.

Testing:

• Load an image containing digits to be recognized.
• Preprocess the image to isolate individual digits.
• Convert each digit into a feature vector (100 pixel values).
• Use the trained classifier to find the nearest match for each feature vector and display the corresponding digit.

Example Code:

import numpy as np
import cv2

# Load data
samples = np.loadtxt('my_letter_recognition.data', np.float32, delimiter=',', converters={ 0 : lambda ch : ord(ch)-ord('A') })
responses = a[:,0]

# Create classifier
model = cv2.KNearest()
model.train(samples, responses)

# Load test image
test_img = cv2.imread('test_digits.png')

# Preprocess image
gray = cv2.cvtColor(test_img, cv2.COLOR_BGR2GRAY)
thresh = cv2.adaptiveThreshold(gray, 255, 1, 1, 11, 2)

# Extract digits
contours, hierarchy = cv2.findContours(thresh, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
digits = []
for cnt in contours:
    if cv2.contourArea(cnt) > 50:
        [x, y, w, h] = cv2.boundingRect(cnt)
        roi = thresh[y:y+h, x:x+w]
        roismall = cv2.resize(roi, (10, 10))
        digits.append(roismall)

# Recognize digits
results = []
for digit in digits:
    roismall = roismall.reshape((1, 100))
    roismall = np.float32(roismall)
    _, results, _, _ = model.find_nearest(roismall, k=1)
    results = results.ravel()
    results = [chr(int(res) + ord('A')) for res in results]

# Display results
output = cv2.cvtColor(test_img, cv2.COLOR_BGR2RGB)
for (digit, (x, y, w, h)) in zip(results, contours):
    cv2.rectangle(output, (x, y), (x + w, y + h), (0, 255, 0), 2)
    cv2.putText(output, str(digit), (x, y), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)

cv2.imshow('Output', output)
cv2.waitKey(0)

This example uses KNearest for digit recognition, but you can replace it with SVM by creating an SVM classifier instead.

The above is the detailed content of How can I implement a basic digit recognition OCR system in OpenCV-Python using KNearest and SVM algorithms?. For more information, please follow other related articles on the PHP Chinese website!