Big Idea: How Does a Computer Process an Image?
When a computer processes an image, it doesn’t “see” a picture the way humans do — it sees a matrix of pixel values. Each pixel is represented by numbers (such as RGB values) and these numbers must be processed to apply effects, transformations, or analysis. This task involves doing the same operation to millions of pixels, which is where parallel processing becomes incredibly useful.
How Image Processing Works: Step-by-Step
1. Image as Data
Every digital image is stored as a 2D array (or grid) of pixels.
Each pixel is made up of three numbers: one for red, one for green, and one for blue — the RGB model.
For example, a 1920×1080 image has 2,073,600 pixels. That’s over 6 million color values (3 per pixel). In the example below, you see a 4x6 image.
2. Processing the Pixels
Let’s say you want to apply a filter that increases brightness. This means:
- For every pixel,
- You add a constant value (e.g., +20) to each RGB component.
For a large image, this operation must be done millions of times, once for each pixel. Doing this sequentially — one pixel at a time — would be slow.
Parallel Processing: The Key Insight
Here’s the accessible example:
Imagine painting a giant mural. If only one person paints (like a CPU), it takes days.
If 1,000 people each paint a small section at the same time (like a GPU), it takes minutes.
That’s exactly what happens in parallel image processing:
- Each GPU core can work on a different pixel at the same time.
- Instead of waiting for one pixel to finish before starting the next, thousands of pixels are processed simultaneously.
- This makes tasks like applying filters, resizing images, or running edge-detection extremely fast.
Real-World Example: Blur Filter
To blur an image:
- Each pixel is replaced by the average of its neighbors.
- This calculation is independent — the new value of pixel (x, y) doesn’t rely on the result of pixel (x+1, y).
- So, each pixel can be processed in parallel, using a simple mathematical function applied over its 3x3 grid.
This is a perfect example of data parallelism — many pieces of data (pixels) can be handled using the same instruction, in parallel.
Why the GPU Is Ideal for This
GPUs are built for Single Instruction, Multiple Data (SIMD) execution:
- Apply one instruction (like "add 20 to the red value")
- To many pieces of data (all pixels)
This is fundamentally why GPUs are better than CPUs for image processing, machine learning, and simulations — because all those tasks involve repeating the same computation many times on independent chunks of data.
Summary
A computer processes images by treating them as large grids of numbers. Image processing means transforming these numbers — and since each transformation is typically independent, parallel processing is ideal. A GPU uses its many cores to do the same operation to many pixels at once, making it far faster than a CPU for this kind of task.