What is an image?

This article attempts to explain one of the most basic concepts of image analysis, namely, what actually is an image.

To humans who learn from birth to use their eyes, along with their other senses, to learn about the world around them, the fact that image analysis seems rather complicated can be confusing. As humans, we can look at an image and instantly make interpretations of the content of that image. We can look at family pictures and recognize individuals. We might even look at family photos of relatives we might never have met and recognize familiar features. We can also look is micrographs and recognize cells, or individual cell components, instantly recognize relationships and differences. But these abilities are woven with several years of learning related to concepts that overlayed on top of other concepts that only exist in a human brain. Many advances in artificial intelligence, and especially Deep Learning seek to duplicate this process to simplify image analysis, but even for this task, it is important to understand what an image actually IS.

To understand what an image is, perhaps it is best to start with what an image is not.

To paraphrase Rene Magritte, these are not cell nuclei:


This is an image of cell nuclei. More specifically, it is a matrix of intensity measurements of the light emitted by fluorophores, projected through a microscope lens onto a light-sensitive wafer of silicon, stored in a computer as binary numbers, and displayed as pixels of various intensity on the computer screen that you are looking at right now.

If we zoom in on one of these "nuclei", we can see more clearly that it is nothing like the actual object it represents:

An image then is a collection of pixels (short for Picture-Element). But as far as the computer is concerned, each pixel is just a number:

The higher the number, the brighter the pixel is displayed on the screen.

So now we can see that *Image* analysis is actually *matrix* analysis. In many ways, the first step in image analysis is displaying these intensity values on the screen in a way that humans can interpret them, but the process of identifying individual objects so that we can measure their features is a process called segmentation. To learn about segmentation, please check this article.