Learn the fundamentals of image interpretation
Explore and make sense of satellite and aerial imagery.
What is image interpretation?
An image interpreter examines aerial or satellite imagery for the purpose of making sense of it, identifying the features it contains, and assessing their significance. Image interpretation techniques were developed progressively over more than 100 years, originally focusing on military applications and later extending to many different applications for scientific and commercial use.
Image analysts rely on strong image interpretation skills.
While modern imagery analysis relies more and more on automated information extraction methods, having good image interpretation skills is an essential foundation. As an image analyst, understanding what the imagery represents will inform and guide every step of your work. For instance, it is virtually impossible to create good training samples or to validate the results of an automated process without good interpretive skills. Understanding what cues your brain used to make sense of the imagery is also crucial when developing your automated workflows and deciding on the specific methods to apply.
This lesson introduces basic imagery interpretation elements and skills.
A set of skills
Successful interpreters must bring skills and knowledge together:
- They have adapted their vision system and trained it to view the world differently.
- They understand how different types of imagery, with varying quality levels, have different potentials for image interpretation.
- They bring their personal knowledge of the world as an additional source of information to fully make sense of what they see in the imagery.
Interpretation of remotely sensed imagery requires new skills.
You'll now learn about the three types of imagery you'll encounter in this lesson.
Three types of imagery
There are many image types, ranging from imagery that is sensitive to the visible and near infrared (NIR) light, through thermal infrared, and into the radio portion of the electromagnetic spectrum.
In this lesson, you'll focus on three common types of images: natural color, panchromatic, and color infrared (CIR).
Three common imagery types
Natural color imagery
Natural color imagery, also called true color, is the most natural for people to look at. It mimics what you see with your own vision system. Color imagery consists of three spectral bands: blue, green, and red. These are displayed in the blue, green, and red channels of a display to form a natural and intuitive image for interpretation.
Here is an example of natural color imagery showing a variety of housing, recreational, agricultural, and commercial examples.
Can you guess what some of the features displayed are? And can you spot the golf course?
Panchromatic imagery
The term panchromatic translates literally to an image that spans the chromatic spectrum. A panchromatic image is a single band image with a spectral band width that is very wide, encompassing the visible light spectrum and sometimes part of the NIR. The black and white appearance of a panchromatic image has an old feel to it (like a vintage black and white picture). However, this is still a very relevant imagery type for remote sensing, since it often has the best clarity for feature data.
Observe how the features change in the example image.
Color infrared imagery
Color infrared (CIR) looks somewhat unusual at first, but it is a common imagery rendering, and it is easy to become accustomed to it.
It includes the near infrared, red, and green spectral bands. Because the human eye cannot see the near infrared light, it is displayed through the red channel. As a result, the display of the other two bands is shifted: the red band is displayed through the green channel, and the green band through the blue channel, as shown in the table below.
This results in an image where healthy vegetation is highlighted and takes on a red tone. Other objects may also shift slightly from their natural color, for instance, taking on a greenish tint.
Observe how the features change on the example image. Can you see examples of vegetation highlighted in red?
Next, you'll learn about the important concept of ground sample distance.
Ground sample distance (GSD)
What is the GSD?
A digital image is made up of pixels in a raster format. For remote sensing, each pixel represents a discrete place on the ground. Each pixel is a sample of the light at that point or place on the ground. It is common to talk about an image that has 6-inch pixels or 3-meter pixels, or some other pixel-related dimension. What this is referring to is the GSD. The GSD is the distance in ground units between the pixel centers.
The GSD is the distance between the pixel centers.
Effects of the GSD
The GSD of an image greatly affects the way it looks and the type of information that can be derived from it. This means that the GSD has a strong impact on image interpretation. The figure below illustrates a vehicle at three different GSDs.
Effects of the GSD on image interpretation
At the far left, the GSD is large, and though the car is detectable as something red, it generally cannot be identified as a car. On the far right, the GSD is small and there are many pixels covering the car. It is easy to identify it as a car. The image in the middle shows a medium GSD. Viewers sometimes correctly identify the car in the middle image, but not always. The number of pixels on the object is the single most important factor for determining if an object is visible at all, identifiable, or merely detectable.
Can you identify the car?
This series of CIR images shows the same location captured at different altitudes. From one image to the next, you can see the effect that changing GSD has on the features that are imaged.
- Try to identify the car in each image. When does it become difficult? Or impossible?
- Can you also identify other elements?
Note: These four images were created to test a sensor's quality. It used test targets that were arranged on the ground. You can see them on the left side of this first image.
Answer
In this first image, the car is rather obvious. So are the color and gray targets. Since this series is CIR imagery, it is easy to identify the bushy vegetation in front and to the side of the vehicle.
As the GSD gets larger, the vehicle is still apparent. The soil that the car is park on appears green and contrasts clearly with the vegetation displayed in red.
Can you identify the rows of red vegetation to the right and left side of the image? Or do you need a wider view to help?
In this view, the car is barely identifiable, but the forest cover in the area now appears clearly. Also it is becoming apparent that there are agricultural fields in the vicinity.
Can you identify the thick dark band that crosses the area from top to bottom?
In this final view, the mix of agriculture and forest is apparent. It is also clear that the dark band mentioned earlier was, in fact, a road in the shade. However, the test targets and the vehicle are now barely detectable. If you were not aware of where the vehicle was parked, you would likely not differentiate it at all from the soil it is parked on.
Next, you'll learn about the seven fundamental cues used in image interpretation.
Seven cues for imagery readout
In your day-to-day life, there are many cues that your vision system uses, allowing you to understand things that you see. These basic cue concepts remain the same when you interpret remotely sensed imagery. However, you must learn how they present themselves in the perspectives that you see in remote sensing. This section will reintroduce these cues and explain their relevance in recognizing objects or observables in the imagery. The number of cues proposed varies slightly among authors, but you will use seven in this lesson.
The seven cues are: shape, size, pattern, texture, tone, shadow, and context.
Shape
Shape is a critical cue for almost any object you want to identify. This example is shown because it illustrates how the shape of the road clearly indicates its function.
Example 1: What is this road?
In this panchromatic image, can the road's shape help you identify its function?
The feature is at the edge of the image, preventing other cues, but the shape of the road is enough of a cue alone to determine its function.
Answer :
The image is, of course, a clover leaf entrance or exit to a limited access highway. If you have ever driven on a limited access highway, the looping curve is familiar to you. The road appears to be going nowhere, but in reality, it is a mechanism to turn onto another road, without the need to stop or yield to conflicting traffic.
Example 2: What is this road?
This next road example also relies on shape for identification. The two gently curving roads merge together, but what are they used for?
Answer
The answer is that they are railroads. Though the GSD is too large to allow confident identification as a railroad (you cannot fully resolve the rails or ties), the gentle curves of the road, known as spiral curves, are required for railroads. Automobiles can transition from straight paths to circular paths without mishap, because they can mitigate the abrupt shift of acceleration by moving from side to side freely in the lane. However, a train is forced to remain rigidly on the rails. So, to avoid the point accelerations, spiral curves are used. This is a clear cue that it is a railroad.
If you are not familiar with spiral curves, here is a diagram, comparing a spiral curve to a circular curve.
Takeaway
Shape and our knowledge of roads and railways help you identify these features.
Size
Some objects are easily categorized and recognized by their size. For instance, most homes are smaller than barns or industrial buildings. Automobiles are typically smaller than trucks. When determining what an object is, size can often be a discriminating attribute.
Which buildings are houses?
Can you use size cues to find out?
Answer
This color image is rather low quality due to JPEG artifacts that result from an image compression that was too aggressive. However, the relative size of the buildings are apparent. You can spot larger buildings, probably barns or other industrial buildings. Houses are smaller and generally have direct access to the roads. Barns and large farm buildings tend to be more than twice the size of a house, and are often offset from both the houses and the road.
Another possible approach would be to measure the absolute size of the buildings, which you can do in a GIS, such as ArcGIS Online or ArcGIS Pro, as they offers measuring tools.
Pattern
The pattern of objects on the ground will often provide insight to their use, their function, or in some cases, their history. As cultures, technology, and styles evolve, there are patterns that emerge on the landscape that are indicative of the moment in time when they originated. The patterns of the objects show when those objects were built, deposited, or grown. The cues of pattern are very important.
Which housing development is newer?
Can some patterns in the images help you answer this question?
Answer
Housing developments in the mid-20th century tended to be linear in pattern, as illustrated in the bottom image. However, as landscape design evolved in the latter part of the 20th century, a more curvilinear pattern of streets evolved as the more typical pattern for subdivisions. The top image is the newer housing development.
Takeaway
Patterns are part of the cultural, functional, and environmental aspects of feature data.
Texture
Texture is an attribute of a region on the image that provides an impression of smoothness or roughness. Textures are very fine patterns that provide cues to the type of feature being observed.
Note that what appears as a distinctive texture at a small GSD can blur and become undistinguishable at a large GSD. So it is important to take the GSD into account for the texture cue.
What are these areas?
Can texture and tone help you identify areas A and B?
Answer
This image was collected in early spring, when deciduous trees have not yet leafed out. In this landscape, there is a mix of agriculture, forested plots, and transitional areas where there is a regrowth of the forest underway. Area A is identified as an agricultural plot, not only because of its shape but because of its smooth texture. The rough texture of area B makes it possible to identify that area as forested.
Tone (and spectral signature)
Tone, a traditional term, refers to a consistent gray level or color that an area in the image has. This cue is relevant for visual interpretation, but especially for sensors that only collect panchromatic data or three spectral bands, such as a simple color camera.
With imagery that offers multiple bands across the electromagnetic spectrum, it is more valuable to think of the spectral signature of an object. Spectral signatures are a manifestation of how light is reflected from an object, across the electromagnetic spectrum. Below, you see an example of how the WorldView-3 sensor sees the spectral signatures of three things: water, maple leaves, and lodge pole pine needles. Note that each one is distinctive. In the visible portion of the spectrum (0.45–0.70) there are already small distinctions, but as you look into the infrared, the distinctions are much more obvious. Understanding spectral signatures is critical to understanding remote sensing.
Softwood versus hardwood
This spectral signature graph shows that hardwood trees like maple have much higher reflectance values than softwood trees like pine.
This is especially true in the Near Infrared (NIR) band. This means that in imagery with a band combination that uses NIR, hardwood appears as much brighter than softwood.
In this CIR, what are the types of wood in areas A and B?
Can you use what you know about the spectral signature of these types of wood to identify them?
Answer
The A area is much brighter, and the B area is darker. Based on the typical spectral signatures you saw earlier, A is hardwood and B softwood.
This type of identification is very useful for forestry applications.
Shadow
Shadows are critical when interpreting objects that stand up from the landscape. For objects like poles, or other narrow, tall things, it is not unusual that the only clear indicator of shape or size is revealed by looking at the shadow that was cast. The image below illustrates the importance of this cue with power transmission towers and distribution poles.
How many power poles are there in this image?
Can you use shadow as a clue?
Answer
There are two types of towers in this image. The very large transmission tower, consisting of two vertical poles and topped with a horizontal bar, is apparent near the middle of the scene. Though it is challenging to see the actual tower, the shadow provides a clear understanding of the object. There are also five smaller distribution poles in the image, visible only because of their shadow. You can also see the three high tension wires as well as their shadows.
Takeaway
Shadow is important:
- Tall features are often more readily detected or identified by shadow.
- Shadow may be used for measurements.
- Shadows can assist in understanding the characteristics of the terrain.
Context
Context is one of the most important tools for the interpreter. This is what provides the interpreter the means to move from object readout toward true interpretation. The term context spans many things, such as:
- The interrelationship of the objects in the image (for instance, trees make up the forest)
- The functional association of various features (for instance, a barn, a house, an agricultural field make a farm)
- The situational aspects of events in the landscape that are not captured in the image but may be available to the analyst elsewhere (for instance, knowing that an image was taken on the morning of a marathon race, allows you to identify the aid stations along a road).
Relevant contextual information can pertain to a variety of topics:
- Slope, aspect, elevation
- Critical infrastructure
- Lines of communication
- Cultural and human geography
- Habitat categories
- Weather and climatic data
- Temporal information
- Pattern of life
In a GIS project, many layers can supplement the imagery, for instance, vector data showing known landmarks or water bodies. These layers can be a good source of contextual information that will help better interpret the imagery.
What is this building?
Can any contextual clues help you find out?
Answer
The building in this image is a school. The associated objects, such as several athletic fields, are indicators of this interpretation.
Takeaway
In this example, the interrelationship of the objects allows for a richer functional understanding of what the buildings in the image are.
From readout to interpretation
This lesson has reviewed the basic cues that you can use to identify objects in an image. When you begin to learn interpretation skills, you often consciously refer to individual cues to reason through what you are seeing. As your skills evolve, you naturally tend to use all of the cues together to recognize objects. It won't take long before you have retrained your vision to be capable of rapidly reading out the objects. Note that you may need to do this with each new sensor type, GSD range, or collection condition you encounter.
Image interpretation is more than reading out objects. It is the skill to recognize the activities that are taking place on the landscape. To achieve this, you must also become adept at deductive reasoning. You will learn to use the collective readout of objects and begin to see them together as indicators of activities. In this way, you'll reach richer levels of understanding of what the image represents. This will be covered in another lesson coming soon.
Summary
In this lesson, you learned about three types of imagery: natural color, panchromatic, and color infrared. You learned about the importance of the GSD for imagery interpretation, and the seven types of cues used by the image interpreter: shape, size, pattern, texture, tone (and spectral signature), shadow, and context.
Having good image interpretation skills is an essential foundation. Understanding what the imagery represents will inform and guide your work as you learn to apply advanced analysis techniques and develop automated imagery processing workflows.