Session 01🎨

Image Fundamentals

Introduction to digital images: color spaces, channel splitting, sampling and quantization.

Learning Objectives

✓Understand BGR, RGB, Grayscale and Binary color spaces
✓Split and visualize individual color channels
✓Apply sampling and quantization to control bit depth

Color Spaces

OpenCV reads images in BGR order (not RGB). You must convert to RGB before displaying with matplotlib. Grayscale reduces three channels to one luminance channel. Binary thresholds the grayscale image at a cutoff (e.g. 127) to produce pure black-and-white.

Ex1_1 — Display image in different formatsPython

import cv2
import numpy as np
import matplotlib.pyplot as plt

Img = cv2.imread('forest.jpg')
Img_RGB   = cv2.cvtColor(Img, cv2.COLOR_BGRA2RGB)
Img_Gray  = cv2.cvtColor(Img_RGB, cv2.COLOR_RGB2GRAY)
_, Img_Binary = cv2.threshold(Img_Gray, 127, 255, cv2.THRESH_BINARY)

plt.subplot(2,2,1), plt.imshow(Img),           plt.title('BGR'),       plt.axis('off')
plt.subplot(2,2,2), plt.imshow(Img_RGB),        plt.title('RGB'),       plt.axis('off')
plt.subplot(2,2,3), plt.imshow(Img_Gray,  cmap='gray'), plt.title('Grayscale'), plt.axis('off')
plt.subplot(2,2,4), plt.imshow(Img_Binary,cmap='gray'), plt.title('Binary'),    plt.axis('off')
plt.tight_layout()
plt.show()

Color Channel Splitting

An RGB image has three channels: Red, Green, and Blue. cv2.split() separates them. To display a single channel in color, merge it with zero-filled arrays for the other two channels.

Ex1_2 — Isolate and display individual channelsPython

B, G, R = cv2.split(Img)
zeros = np.zeros_like(B)

B_color = cv2.merge([B, zeros, zeros])
G_color = cv2.merge([zeros, G, zeros])
R_color = cv2.merge([zeros, zeros, R])

plt.subplot(2,2,1), plt.imshow(Img_RGB),  plt.title('RGB'),   plt.axis('off')
plt.subplot(2,2,2), plt.imshow(B_color),  plt.title('Blue'),  plt.axis('off')
plt.subplot(2,2,3), plt.imshow(G_color),  plt.title('Green'), plt.axis('off')
plt.subplot(2,2,4), plt.imshow(R_color),  plt.title('Red'),   plt.axis('off')
plt.tight_layout()
plt.show()

Sampling & Quantization

Quantization reduces the number of intensity levels. An $n$ -bit image has $2^{n}$ levels — a 1-bit image is pure black-and-white ( $2$ levels), while an 8-bit image has $2^{8} = 256$ levels. Each pixel is quantized by rounding to the nearest level gap.

\text{gap} = \frac{256}{2^{n}} \qquad\qquad s = \left\lceil \frac{r}{\text{gap}} \right\rceil \cdot \text{gap} - 1

Smaller

n

→ fewer gray levels → visible banding

Ex1_4 — Quantize image at different bit depthsPython

Img = cv2.imread('rose.jpg').astype(np.float64)

for num_bits in range(1, 9):
    num_levels = 2 ** num_bits
    level_gap  = 256 / num_levels
    q_img = np.uint8(np.clip(np.ceil(Img / level_gap) * level_gap - 1, 0, 255))
    plt.subplot(4, 4, num_bits)
    plt.imshow(q_img)
    plt.title(f'{num_bits}-bit')
    plt.axis('off')

plt.tight_layout()
plt.show()

→NextImage Arithmetic