5.17. Digital Image Interpretation Methods

This section deals with the interpretation of optical digital remote sensing images. Digital remote sensing images consist of pixels, and have a square grid structure, where one grid represents a pixel (i.e., the smallest element in a digital image). Each pixel is associated with a DN in a particular wavelength.

Digital images are made up of small squares called pixels.Each pixel is a tiny dot that contains information about a specific color or brightness at that spot. The value associated with a pixel, known as the Digital Number (DN), represents its intensity at different wavelengths of light (like red, green, blue) which are captured by sensors on satellites or other imaging devices. This structure allows for precise analysis and interpretation of images.

Think of a digital image like a quilt made up of many small squares (the pixels). Each square is a different color and together they form a larger picture. Just as each piece of fabric has a certain color or pattern, each pixel has a DN value that indicates how it looks in the overall image.

Identification and separation of objects/features with respect to their DN values is called digital image processing. Pixels with similar DN values are grouped into various classes.

Digital image processing refers to the use of algorithms and software to manipulate and analyze images. It involves identifying objects or features within an image by examining the DN values of the pixels. Pixels that have similar DN values can indicate similar objects or features in the image, which allows for their grouping into categories. This classifying method helps in analyzing large datasets efficiently.

Imagine sorting a box of mixed candies based on their colors and characteristics. Each group of similarly colored candies represents a class. In digital image processing, similar DN values help group pixels that belong to the same visual category, making it easier to understand the content of the image.

A comparison between visual and digital methods of interpretation is given in Table 5.8.

Visual interpretation relies on human judgment to analyze satellite images, requiring physical copies of images and simpler tools. In contrast, digital interpretation leverages advanced software to automatically analyze multiple images simultaneously, often yielding more consistent and quantitative results. This comparison highlights the strengths and weaknesses of both approaches to aid in image analysis.

Consider how a chef can taste ingredients and create a recipe (visual interpretation) versus using a food processor that can swiftly combine and measure ingredients accurately (digital interpretation). Both have their places, but the food processor may yield consistent results faster, just as digital methods often produce more reliable data.

Image pre-processing involves the initial processing of raw image data to apply corrections for geometric distortions, calibrate the data radiometrically, and remove the noise present in the data, if any.

Pre-processing is a crucial step in preparing raw image data for analysis. It includes correcting any distortions caused by the way the image was captured (geometric corrections), adjusting for inaccuracies in how brightness values are recorded (radiometric calibration), and filtering out unwanted noise that may interfere with analysis. These steps enhance the quality of the final image and ensure that interpretation is accurate.

Before painting a wall, you would typically clean it, patch holes, and add a primer coat to ensure the paint adheres well and the surface looks good. Similarly, image pre-processing 'prepares' the raw image data by correcting and enhancing it before it’s subjected to detailed analysis.

Geometric Corrections involve two basic steps: georeferencing and resampling.

Geometric corrections are essential in accurately aligning the image with the ground coordinates. Georeferencing is the process of converting image coordinates into real-world coordinates, correcting distortions like those caused by the Earth's curvature or sensor movement. Resampling adjusts the positions of the pixels in the corrected image to make sure they fit properly in the new coordinate system, using different interpolation methods.

Imagine taking a photo while tilting your camera. When you straighten that photo to align with the horizon, you’re effectively georeferencing it. Resampling would be like adjusting the pixels so they don't get stretched or squashed in the process, retaining the quality of the image much like a well-stretched canvas.

Atmospheric correction is done to modify the DN values to remove noise, i.e., contributions to the DN due to intervening atmosphere.

Atmospheric correction seeks to address distortions in the data caused by the atmosphere through which the satellite data is captured. Factors such as haze can alter the recorded DN values, making objects appear differently than they truly are. This correction uses methods, like dark object subtraction, where the lowest DN value is identified and subtracted from all pixel values to eliminate haze effects.

Think of trying to see through a dirty window. The grime can obscure your view. Cleaning the glass or adjusting how you look through it can give you a clearer view. Just like that, atmospheric corrections help clarify the data captured by remote sensing, allowing us to see the true characteristics of the landscape.

Image enhancement is mainly carried out to improve the quality of images. Often, the images do not have the optimum contrast so that objects are not clearly identified visually.

Image enhancement techniques modify images to improve their clarity and contrast, making it easier to distinguish features. These methods manipulate the visual appearance of the image, without adding any new information. Techniques often involve adjusting the brightness and contrast levels to highlight important details that may be obscured.

Enhancing an image is like increasing the brightness on your TV screen when you can't see the details well. By adjusting the settings on the TV, you get a clearer picture. Image enhancement works on photos and satellite images in a similar way, ensuring that features stand out for better analysis.

Contrast enhancement involves changing the original DN values so that more available range is utilized, thereby increasing the contrast between the objects and their background.

Contrast enhancement is a specific image enhancement technique that adjusts the pixel values to expand the range of tones in the image. By identifying minimum and maximum DN values, the existing values can be stretched to utilize the entire available range. This makes objects more distinguishable from their backgrounds, especially in low-contrast images.

Picture a drawing done in pencil where you can barely see the faint lines. Darkening the lines gives it more contrast, making each element clearer. That's exactly what contrast enhancement does to images—it sharpens the details so that they can be easily identified.

Digital image classification is a software-based classification technique used for information extraction from optical images based on their DN values.

Image classification involves grouping pixels based on their spectral characteristics. This is accomplished using statistical algorithms that assign each pixel to specific classes based on similarity in DN values. This process can be supervised, where analysts provide training data, or unsupervised, where the algorithm determines the classifications independently.

Think of a classroom where a teacher organizes students into groups based on their interests (supervised classification). In contrast, a party where guests naturally form groups based on who they talk to (unsupervised classification). Both methods group similar items together based on defined or emerging characteristics.

Accuracy assessment is a critical step in the use of results from analyses based on remotely sensed data to evaluate the classification quality.

After classifying images, it’s essential to verify how accurate the classification is. Accuracy assessment involves comparing the classified image against a 'truth' reference (known data) to see how many pixels were correctly classified. This analysis helps identify errors and improve the classification process in the future.

Imagine taking a test at school where you compare your answers with the correct ones. This helps you see where you might have made mistakes. Accuracy assessment works the same way for satellite images, ensuring researchers understand how well their classifications match the real world.

--