Exploring Images as Data

Step 1: Setup Your Environment

Create a new Python file or Jupyter notebook and import the necessary libraries:

import numpy as np
from PIL import Image
import matplotlib.pyplot as plt

# Optional: for more advanced operations
import cv2

Download a sample image for this exercise, or use any image from your computer. Save it as sample.jpg in your working directory.

Step 2: Load and Inspect an Image

Load an image and examine its basic properties:

# Load the image
img = Image.open("sample.jpg")

# Display basic information
print(f"Format: {img.format}")
print(f"Mode: {img.mode}")
print(f"Size: {img.size}")  # (width, height)

# Convert to numpy array for numerical analysis
img_array = np.array(img)

print(f"\nArray shape: {img_array.shape}")
print(f"Data type: {img_array.dtype}")
print(f"Min value: {img_array.min()}")
print(f"Max value: {img_array.max()}")

Step 3: Explore Pixel Values

Access and understand individual pixel values:

# Access a single pixel (row 100, column 150)
pixel = img_array[100, 150]
print(f"Pixel at (100, 150): {pixel}")
print(f"Red: {pixel[0]}, Green: {pixel[1]}, Blue: {pixel[2]}")

# Extract color channels
red_channel = img_array[:, :, 0]
green_channel = img_array[:, :, 1]
blue_channel = img_array[:, :, 2]

print(f"\nRed channel shape: {red_channel.shape}")
print(f"Red channel mean: {red_channel.mean():.2f}")

Step 4: Visualize the Image and Channels

Create visualizations to understand the image structure:

# Create a figure with subplots
fig, axes = plt.subplots(2, 2, figsize=(10, 10))

# Original image
axes[0, 0].imshow(img_array)
axes[0, 0].set_title("Original Image")
axes[0, 0].axis("off")

# Red channel
axes[0, 1].imshow(red_channel, cmap="Reds")
axes[0, 1].set_title("Red Channel")
axes[0, 1].axis("off")

# Green channel
axes[1, 0].imshow(green_channel, cmap="Greens")
axes[1, 0].set_title("Green Channel")
axes[1, 0].axis("off")

# Blue channel
axes[1, 1].imshow(blue_channel, cmap="Blues")
axes[1, 1].set_title("Blue Channel")
axes[1, 1].axis("off")

plt.tight_layout()
plt.savefig("channel_visualization.png")
plt.show()

Step 5: Basic Image Manipulations

Perform common preprocessing operations:

# Resize the image
resized = img.resize((224, 224))
print(f"Resized shape: {np.array(resized).shape}")

# Convert to grayscale
grayscale = img.convert("L")
gray_array = np.array(grayscale)
print(f"Grayscale shape: {gray_array.shape}")

# Crop a region (left, top, right, bottom)
cropped = img.crop((100, 100, 300, 300))

# Rotate
rotated = img.rotate(45)

# Visualize transformations
fig, axes = plt.subplots(2, 2, figsize=(10, 10))

axes[0, 0].imshow(np.array(resized))
axes[0, 0].set_title("Resized (224x224)")

axes[0, 1].imshow(gray_array, cmap="gray")
axes[0, 1].set_title("Grayscale")

axes[1, 0].imshow(np.array(cropped))
axes[1, 0].set_title("Cropped")

axes[1, 1].imshow(np.array(rotated))
axes[1, 1].set_title("Rotated 45°")

for ax in axes.flat:
    ax.axis("off")

plt.tight_layout()
plt.savefig("transformations.png")
plt.show()

Step 6: Normalize Pixel Values

Learn to normalize images for machine learning:

# Standard normalization (0-1 range)
normalized = img_array.astype(np.float32) / 255.0
print(f"Normalized range: [{normalized.min():.2f}, {normalized.max():.2f}]")

# ImageNet normalization (commonly used for pre-trained models)
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])

imagenet_normalized = (normalized - mean) / std
print(f"ImageNet normalized range: [{imagenet_normalized.min():.2f}, {imagenet_normalized.max():.2f}]")

Step 7: Calculate Image Statistics

Compute useful statistics about your image:

# Per-channel statistics
for i, color in enumerate(["Red", "Green", "Blue"]):
    channel = img_array[:, :, i]
    print(f"{color}: mean={channel.mean():.1f}, std={channel.std():.1f}")

# Histogram of pixel values
fig, axes = plt.subplots(1, 3, figsize=(12, 4))
colors = ["red", "green", "blue"]

for i, (ax, color) in enumerate(zip(axes, colors)):
    ax.hist(img_array[:, :, i].ravel(), bins=256, color=color, alpha=0.7)
    ax.set_title(f"{color.capitalize()} Channel Histogram")
    ax.set_xlabel("Pixel Value")
    ax.set_ylabel("Frequency")

plt.tight_layout()
plt.savefig("histograms.png")
plt.show()