Skip to content

4 Data Exploration

Jump to bottom
John E Stranzl Jr edited this page Feb 26, 2026 · 3 revisions

Below are the tools associated with data exploration in GRIME AI:

Color Segmentation

Color Segmentation is a tool for extracting visual information like color, intensity, and texture from a region of interest in an image. The Color Segmentation tool works in tandem with the Image Analysis tab , so both must be opened to extract image features. Before utilizing Color Segmentation, a reference folder must also be chosen in the Data Exploration tab. Below is an example of the Image Analysis tab open with a successful image fetch.


After clicking the Color Segmentation button on the navigation bar, the below window should appear.

This window has several options, which will be broken down into detail, starting with the Define ROI field.

A ROI is a region of interest used for visual analysis. Any number of ROIs can be created to analyze different sections of an image.

Each ROI should be given a name, to be typed into the Name field. Although the names are arbitrary, descriptive names are recommended.

Color Clusters, located to the right of the Name field, are used to find average colors within the ROI. By increasing or decreasing the amount, more or fewer color clusters will be shown.

The Shape field changes the shape of the ROI to be used. Currently there are two options: Rectangle and Ellipse, and one can be chosen per ROI.

Display ROIs displays an outline of the ROI on the left window of the Image Analysis tab. Display ROI Colors displays the colors from the color cluster in the right window of the Image Analysis tab. Delete All ROIs wipes all current ROIs.

Several pieces of visual data can be extracted from a whole image or from a ROI within an image.

Feature Calculations gives all current options for visual feature extraction.

  • Region Select allows for the selection of either the entire image, or a ROI for feature extraction.
  • Feature Select gives a number of options for types of visual data to be extracted.
    1. Shannon Entropy
    2. Intensity
    3. Color (HSV)
    4. Texture
  • Greenness
    1. GCC (Greenness Index)
    2. GLI (Green Leaf Index)
    3. ExG (Excess Green Index)
    4. NDVI (Normalized Difference Vegetation Index)

Build Feature File creates a CSV with all selected features after being analyzed.

Edge Detection

The Edge Detection Methods tool is found on the navigation bar and is used to automatically detect edges of streams and foliage lines in images using a variety of computer vision approaches. The example window can be found below.

Each radio button in the menu corresponds to a different edge detection method.

Canny is the standard edge detection algorithm. It has a high computing cost, but can be more accurate than Sobel. It works by first applying a Gaussian filter to an image, and then finding edges by comparing gradients of pixels in the image.

Three values are adjustable under the Canny edge detection method: HIGH Threshold, Low Threshold, and Order of Kernel.

The process of finding the upper and lower thresholds is called hysteresis. It is also the final process in Canny edge detection.

From the OpenCV official documentation:

  • HIGH Threshold: If a pixel gradient is higher than the upper threshold, the pixel is accepted as an edge.
  • Low Threshold: If a pixel gradient value is below the lower threshold, then it is rejected.
  • Order of Kernel: If the pixel gradient is between the two thresholds, then it will be accepted only if it is connected to a pixel that is above the upper threshold.

An upper:lower ratio between 2:1 and 3:1 is recommended.

See here for more information on Canny.

Sobel is the second built-in edge detection method. It works by calculating the edges within an image as derivatives.

It comes with 4 editable options (which correspond to direction of the derivative): Sobel X, Sobel Y, Sobel XY, and Kernel Size.

  • Sobel X scans the image on the X-axis, and is concerned with horizontal changes in the pixel gradient.
  • Sobel Y scans the image on the Y-axis, and is concerned with vertical changes in the pixel gradient.
  • Sobel XY scans the image on a diagonal axis, and is concerned with diagonal changes in the pixel gradient.

Kernel size can be set to 1, 3, 5, or 7. A change in the kernel size corresponds to a change in the matrix size used in the differentiation operation.

See here for more information on Sobel.

Three additional edge detection methods are featured in GRIME-AI: Laplacian, SIFT, and ORB.

Laplacian is a derivative of the Sobel derivative. It uses one symmetrical kernel, whereas Sobel uses X and Y. See here for more information.

SIFT: For more information on Sift, see this paper.

ORB is an unpatented alternative to SIFT. See here for more information.

Water Level Detection

Feature coming soon.


Back to Top

For updates on GRIME and blog-style content, visit: gaugecam.org

Clone this wiki locally