Image Array Layout & Memory Structure

🧠 Why Image Array Structure Matters

In computer vision, bugs often come not from algorithms,
but from wrong assumptions about image memory layout.

Understanding array structure is essential for:

• Correct pixel access
• Fast processing
• Writing your own image loader
• Debugging strange artifacts

---

🧱 Image as Raw Memory

An image in memory is a 1D byte array, interpreted as 2D or 3D.

Conceptually:

memory[]  →  rows  →  pixels  →  channels

But rows are not always tightly packed.

---

📐 Basic Image Dimensions

Symbol	Meaning
W	Width (pixels)
H	Height (pixels)
C	Channels
B	Bytes per channel
RowBytes	Actual bytes per row
Stride	Distance between row starts

---

📏 Stride (Pitch)

Stride = number of bytes between the start of two consecutive rows.

row y starts at: base + y * stride

Stride ≥ (W × C × B)

Why larger?

• Alignment
• Padding
• SIMD / GPU constraints

---

🧩 Padding

Padding = unused bytes at the end of each row.

Purpose:

• Memory alignment
• Faster access
• Hardware constraints

Padding bytes do not represent pixels.

---

🟦 Bitmap (BMP) Row Alignment Rule

The Rule

Each row in a BMP file must be aligned to 4-byte boundaries.

Meaning:

RowBytes = ceil((W × bitsPerPixel) / 32) × 4

---

Example: 24-bit BMP

• Width = 3 pixels
• 24 bits per pixel = 3 bytes

Raw row size:

3 × 3 = 9 bytes

BMP requires multiple of 4:

Padding = 3 bytes
RowBytes = 12 bytes

---

Visual Layout

[B G R][B G R][B G R][P][P][P]

P = padding byte

---

🔄 Bottom-Up Storage (BMP)

Most BMP files store rows bottom to top.

Memory order:

Last row
...
First row

If ignored:

• Image appears vertically flipped

---

🎨 Channel Interleaving

Interleaved (HWC)

[R G B][R G B][R G B]

Planar (CHW)

RRR...
GGG...
BBB...

BMP / PNG / JPEG files:

• Usually interleaved

Deep learning frameworks:

• Often planar

---

📦 Pixel Size Calculation

Bytes per pixel:

PixelBytes = C × B

Total memory (excluding padding):

H × W × C × B

Actual memory:

H × stride

---

⚠️ Common Bugs

❌ Assuming stride == width × channels

• Fails on BMP
• Fails on aligned buffers

❌ Ignoring bottom-up layout

• Flipped images

❌ Forgetting padding when copying

• Shifted or torn images

❌ Mixing RGB and BGR

• Color swapped output

---

🧪 Safe Pixel Access Formula

To read pixel (x, y):

address = base + y * stride + x * PixelBytes

Never assume contiguous rows.

---

🧠 Practical CV Insight

Most libraries expose:

• width
• height
• channels
• stride (or step)

Always trust stride, not width.

---

📊 Summary Table

Concept	Why It Exists
Stride	Row alignment
Padding	Hardware efficiency
Bottom-up	Legacy BMP
Interleaving	Simplicity
Planar	Vectorization

---

🎯 Takeaway

Images are memory structures first, pictures second.

If you understand:

• Stride
• Padding
• Alignment
• Layout

You can:

• Write loaders
• Avoid silent bugs
• Optimize performance

This knowledge separates CV users from CV engineers 🚀