When performing machine vision debugging, exposure is almost invariably the first hurdle that you can't avoid.

But when judging exposure on the spot, one of the most common mistakes is:

Too dark? Increase the exposure. Too bright? Decrease the exposure.

Sounds fine.

But when actually working on a project, simply looking at whether something “lights up” or not is far from enough.

Because when judging whether an exposure is good or not, what really matters isn't the brightness of the image, but rather:

Is the information about the key area still available?

I. What constitutes proper exposure?

Many people interpret “normal exposure” as:

The screen brightness is moderate and looks quite comfortable.

Actually, in machine vision, it’s not entirely like that.

At the heart of proper exposure lies the need to fully preserve the information being detected.

For example:

· The character strokes can still be distinguished.

· The edges haven't blended into the background.

· The bright areas aren't completely washed out.

· The dark areas still have texture and layers.

· Key features can be stably extracted.

Therefore, even if an image is slightly dark overall, as long as its key features are clear, it can still be suitable for visual inspection.

Conversely, if an image looks bright and clear but has lost its text and edges, it still can’t be considered properly exposed.

Visual images aren't meant to be looked at—they're meant to be detected.

II. How do you determine underexposure? Look to see whether the dark areas have lost their details.

The most obvious manifestation of underexposure is darkness.

But don’t immediately assume underexposure just because the image appears dark.

What we should really focus on is:

Has the useful information in the dark areas disappeared?

For example, black workpieces, shadow areas, and dark-colored characters.

As exposure continues to decrease, these areas will gradually lose their tonal range.

The edges that were once still visible began to blend into the background.

Textures that were once distinguishable gradually merged into a single expanse of black.

At this point, it’s no longer just a matter of “the image being a bit darker.”

But:

The information has been overwhelmed.

Even if you later use software to brighten the image, it’s still difficult to truly restore the original details.

So, a very practical tip for identifying underexposure is:

Don't just look at whether it's dark or not—look to see if there’s anything inside the darkness.

III. How do you determine overexposure? Focus especially on the bright areas.

The same goes for overexposure.

Just because a picture is bright doesn't necessarily mean it's overexposed.

The real danger is:

The highlighted area has lost its layers.

Especially when you encounter these locations on-site, take an extra moment to look closely:

· Metallic reflective surface

· White label

· Highlighted characters

· Strong reflection edge

· Localized bright spot

After exposure is applied, these areas usually encounter problems first.

At first, it was just bright.

As you zoom in further, the local details begin to diminish.

Finally, it directly turned completely white.

At this point, even if the entire image is subsequently darkened, the original characters, edges, and surface textures will not reappear.

Because that information was already gone by the time it was collected.

So, to determine overexposure, don't look at:

“Isn’t this picture really bright?”

But rather, look at:

“Is there still any gradation in the brightest area?”

IV. Why does the image look normal overall, yet individual areas still fail to expose properly?

This is a pitfall that’s particularly easy to fall into during on-site debugging.

Sometimes the entire image looks perfectly fine.

The brightness is also quite uniform.

But if you zoom in on the area that really needs to be inspected, you’ll find:

The characters have begun to turn pale.

The reflective edge has reached its limit.

The silhouette in the shadows is almost gone, too.

That is to say:

Exposure issues don't necessarily appear across the entire image at once; often, problems first arise in specific localized areas.

Especially for highly reflective metals, black plastics, and workpieces with significant contrast between light and dark areas.

The overall image brightness should be used only as a reference.

What truly determines whether an exposure is appropriate is the detection area.

So when adjusting exposure on-site, I’d recommend prioritizing the following areas:

Character region, highlight region, shadow region, and the edges that actually participate in detection.

V. A simple and effective order for adjusting exposure

There’s no need to make the exposure judgment overly complicated on-site.

You can just watch it in this order.

First, look at the highlight.

Is there any region that has turned completely white?

Can characters and edges still be separated?

If the highlights are already blown out, the exposure is likely too high.

Look at the shadows again.

Are there any details left in the shadows and dark areas?

Is the outline starting to blend into the background?

If the dark areas are directly flattened into a single tone, the exposure might be underexposed.

Finally, check the detection area.

This is the most important step.

Are the features still intact in the area that truly needs to be identified, located, and measured?

If it’s complete, it means there’s still available space for exposure.

If the key features are already gone, no matter how beautiful the entire image is, it’s all meaningless.

VI. with normal exposure, there isn't actually a fixed “standard brightness.”

This is an important point when working on on-site projects.

Different targets, different materials, and different inspection tasks all have varying requirements for exposure.

Some projects require looking at the shadows.

Some projects focus on suppressing highlights.

For some projects, as long as the character edges are stable, it doesn't matter if other areas are slightly brighter or darker.

So don’t keep adjusting the exposure back and forth just to make the whole image look “just right.”

Many times, this so-called “just right” is merely what the human eye finds comfortable.

Machine vision should more importantly strive for:

The information needed for the test is just right.

Final summary

Normal exposure, underexposure, and overexposure can actually be judged with one simple sentence:

Normal exposure: The key information is still there.

Underexposure: Details in the dark areas begin to be lost.

Overexposure: Details in the bright areas begin to be lost.

So adjust the exposure on-site—don’t just focus on asking about the entire image.

“Is it bright enough?”

One should rather ask:

Are the characters still there?

Is the edge still there?

Did it reach its peak?

Is the dark area crushed?

What exposure adjustment truly aims to preserve is never a picture that merely “looks pleasing.”

But rather the information that the subsequent algorithm truly needs.

First secure the information, then talk about brightness.

This is often more helpful than simply fixating on the exposure settings themselves.