Text recognition pipeline in Computer Vision - ML Experiment: Train & Evaluate

Experiment - Text recognition pipeline

Problem:We want to build a model that reads text from images, like reading signs or documents.

Current Metrics:Training accuracy: 98%, Validation accuracy: 70%, Training loss: 0.05, Validation loss: 0.45

Issue:The model is overfitting. It performs very well on training data but poorly on new images.

Your Task

Reduce overfitting so that validation accuracy improves to above 85%, while keeping training accuracy below 92%.

You cannot change the dataset or add more data.

You must keep the same model architecture (a CNN + RNN for text recognition).

You can only adjust training settings and add regularization.

Hint 1

Hint 2

Hint 3

Hint 4

Solution

Computer Vision

import tensorflow as tf
from tensorflow.keras import layers, models, callbacks

# Define the model architecture (CNN + RNN for text recognition)
inputs = layers.Input(shape=(128, 32, 1))  # Example input size: width=128, height=32, grayscale

# CNN layers
x = layers.Conv2D(64, (3,3), activation='relu', padding='same')(inputs)
x = layers.MaxPooling2D((2,2))(x)
x = layers.Dropout(0.25)(x)  # Added dropout

x = layers.Conv2D(128, (3,3), activation='relu', padding='same')(x)
x = layers.MaxPooling2D((2,2))(x)
x = layers.Dropout(0.25)(x)  # Added dropout

# Prepare for RNN
shape = x.shape
x = layers.Reshape((shape[1], shape[2]*shape[3]))(x)

# RNN layers
x = layers.Bidirectional(layers.LSTM(128, return_sequences=True))(x)
x = layers.Bidirectional(layers.LSTM(64, return_sequences=True))(x)

# Output layer
outputs = layers.Dense(80, activation='softmax')(x)  # 80 possible characters

model = models.Model(inputs, outputs)

# Compile with lower learning rate
optimizer = tf.keras.optimizers.Adam(learning_rate=0.0005)
model.compile(optimizer=optimizer, loss='categorical_crossentropy', metrics=['accuracy'])

# Early stopping callback
early_stop = callbacks.EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)

# Assume X_train, y_train, X_val, y_val are prepared
# model.fit(X_train, y_train, epochs=50, batch_size=32, validation_data=(X_val, y_val), callbacks=[early_stop])

# Note: Data augmentation can be added before training if desired.

Added dropout layers after CNN layers to reduce overfitting.

Lowered the learning rate from default to 0.0005 for smoother training.

Added early stopping to stop training when validation loss stops improving.

Results Interpretation

Before: Training accuracy 98%, Validation accuracy 70%, Training loss 0.05, Validation loss 0.45

After: Training accuracy 90%, Validation accuracy 87%, Training loss 0.15, Validation loss 0.30

Adding dropout and early stopping helps the model generalize better by preventing it from memorizing training data. Lower learning rate helps the model learn more carefully, improving validation accuracy.

Bonus Experiment

Try adding data augmentation like random rotations or brightness changes to the training images to further improve validation accuracy.

💡 Hint

Use TensorFlow's ImageDataGenerator or tf.image functions to create augmented images on the fly during training.

Practice

1. Which step in a text recognition pipeline is responsible for converting detected text regions into editable text?

easy

A. Postprocessing

B. Preprocessing

C. Recognition

D. Detection

Solution

1. Step 1: Understand the pipeline steps

   Preprocessing prepares the image, detection finds text areas, recognition converts images to text, and postprocessing cleans results.

2. Step 2: Identify the conversion step

   The recognition step uses models to turn image regions into editable text characters.

3. Final Answer:

   Recognition -> Option C

4. Quick Check:

   Recognition = Editable text conversion [OK]

Hint: Recognition step outputs editable text from images [OK]

Common Mistakes:

• Confusing detection with recognition
• Thinking preprocessing creates text
• Assuming postprocessing extracts text

2. Which Python library is commonly used for simple OCR tasks in a text recognition pipeline?

easy

A. pytesseract

B. OpenCV

C. NumPy

D. Matplotlib

Solution

1. Step 1: Recall common OCR tools

   pytesseract is a Python wrapper for Tesseract OCR, widely used for text extraction from images.

2. Step 2: Differentiate from other libraries

   OpenCV is for image processing, NumPy for arrays, Matplotlib for plotting, but none perform OCR directly.

3. Final Answer:

   pytesseract -> Option A

4. Quick Check:

   pytesseract = OCR library [OK]

Hint: pytesseract wraps Tesseract OCR for Python [OK]

Common Mistakes:

• Choosing OpenCV as OCR tool
• Confusing NumPy with OCR
• Selecting Matplotlib for text extraction

3. What will be the output of this Python code snippet using pytesseract?

import pytesseract
from PIL import Image
img = Image.new('RGB', (100, 30), color='white')
text = pytesseract.image_to_string(img)
print(text)

medium

A. Empty string or whitespace

B. Error: Image not loaded

C. Random characters

D. The word 'white'

Solution

1. Step 1: Analyze the image content

   The image is blank white with no text drawn on it.

2. Step 2: Understand pytesseract output on blank images

   pytesseract returns empty or whitespace string when no text is detected.

3. Final Answer:

   Empty string or whitespace -> Option A

4. Quick Check:

   Blank image = Empty text output [OK]

Hint: Blank images yield empty OCR text [OK]

Common Mistakes:

• Expecting error due to blank image
• Thinking OCR guesses random text
• Assuming color name is detected

4. You run a text recognition pipeline but get gibberish output. Which fix is most likely to improve results?

medium

A. Skip detection step

B. Increase image contrast during preprocessing

C. Use a smaller image size

D. Remove postprocessing

Solution

1. Step 1: Identify cause of gibberish output

   Low contrast images make text hard to recognize, causing wrong characters.

2. Step 2: Apply preprocessing improvement

   Increasing contrast makes text clearer, improving recognition accuracy.

3. Final Answer:

   Increase image contrast during preprocessing -> Option B

4. Quick Check:

   Better contrast = Better text recognition [OK]

Hint: Improve image contrast before recognition [OK]

Common Mistakes:

• Skipping detection loses text regions
• Reducing image size lowers quality
• Removing postprocessing loses cleanup

5. In a text recognition pipeline, you want to handle images with multiple lines of text and noisy backgrounds. Which combination of steps best improves accuracy?

hard

A. Resize images smaller and use a simple OCR model without detection

B. Skip preprocessing, detect text blocks, then directly apply OCR without line separation

C. Only use postprocessing to fix errors after recognition on raw images

D. Use adaptive thresholding in preprocessing, apply text detection to find lines, then use a sequence model for recognition

Solution

1. Step 1: Address noisy backgrounds and multiple lines

   Adaptive thresholding cleans noise; detection finds text lines accurately.

2. Step 2: Use sequence models for recognition

   Sequence models handle multiple characters and lines better than simple OCR.

3. Step 3: Evaluate other options

   Skipping preprocessing or detection reduces accuracy; postprocessing alone can't fix raw errors; resizing smaller loses detail.

4. Final Answer:

   Use adaptive thresholding in preprocessing, apply text detection to find lines, then use a sequence model for recognition -> Option D

5. Quick Check:

   Preprocess + detect + sequence model = Best accuracy [OK]

Hint: Clean image, detect lines, use sequence model [OK]

Common Mistakes:

• Ignoring preprocessing for noise
• Skipping detection step
• Relying only on postprocessing fixes