Automated Drone Image Analysis Tool

Released – May 2019

Synopsis

The color detection algorithm looks at each pixel in an image to determine if it is within a user-specified color range.  Any sufficiently sized areas where all pixels are within the selected color range are flagged as “areas of interest” in the image. 

The color detection algorithm in ADIAT utilizes the inRange function in OpenCV.

Input Parameters

1. Reference Color: The base color we will be searching for in the images.
2. Color Range Selectors (Red, Green, Blue): Allowed deviation from the selected base color for each color component.
   
3. Color Range Min, Mid, Max*: Shows the min, mid, and max color for the selected color rage.
4. Color Range Visualizer: Shows the selected and unselected color range in a HSL color maps.

* The color range is a one-dimensional representation of the three-dimensional selected color range.  It is intended to be a guide and is not comprehensive.

Libraries, Research, and Tutorials

• https://docs.opencv.org/3.4/da/d97/tutorial_threshold_inRange.html
• https://docs.opencv.org/3.4/d2/de8/group__core__array.html#ga48af0ab51e36436c5d04340e036ce981
• https://pyimagesearch.com/2014/08/04/opencv-python-color-detection/

Released – May 2019

Synopsis

The RX (Reed-Xiaoli) anomaly detector extracts targets that are spectrally distinct from the image background by using the squared Mahalanobis distance as a measure of how anomalous a pixel is with respect to an assumed background.  Once the RX scores have been calculated, each scored pixel is evaluated against a user-specified threshold to declare whether or not it should declared anomalous.  Any sufficiently large groups of anomalous pixels are flagged as “areas of interest” in the image.

The RX anomaly detector used in ADIAT utilizes the rx function in Spectral Python.

Note: In previous versions of ADIAT this algorithm was called “Color Anomaly”

Input Parameters

1. Image Segments: Divides the image in to equally-sized segments so that the target pixel will be compared to a section of the image rather than the entire image. 
2. Sensitivity: Threshold for determining if a pixel is anomalous with respect to the background.  Default value is 5

Libraries, Research, and Tutorials

• https://www.spectralpython.net/algorithms.html#rx-anomaly-detector
• http://cver.hrail.crasar.org/algorithm/

Released – October 2023

Synopsis

“When a pattern, perfectly known as a sub-image g, is searched for in an image f, then the cross-correlation between g and f is a very efficient technique. This technique is often known as matched filter” (Vincent Mazet).

As applied here, rather than a sub-image (spatial domain), we are looking for groups of pixels that correlate to a specified color (spectral domain).  Individual pixels are scored based on how close they are to the reference color and then the scores are evaluated against the user-provided threshold. Any sufficiently large groups of pixels scored above the threshold are flagged as “areas of interest” in the image.

The matched filter detector used in ADIAT utilizes the matched_filter function in Spectral Python.

Input Parameters

1. Reference Color: The color we will be searching for in the images.
2. Threshold: Sets the minimum “score” (on a scale of .1 – 1) that we will consider to be a match to the reference color once the matched filter detector is run.  Default value is .3
3. Color Range Visualizer: Shows the selected and unselected color range in a HSL color maps.

Libraries, Research, and Tutorials

• https://www.spectralpython.net/algorithms.html#matched-filter
• https://courses.cs.washington.edu/courses/cse591n/07sp/papers/14_1hyperspectralprocessing.pdf
• https://vincmazet.github.io/bip/detection/matched-filter.html

Released – October 2023

Synopsis

Lighting conditions can have a dramatic impact on the representation of colors in an image which, in turn, means that color identification can be problematic with UAV photos when they are taken at different times of day or with variable cloud cover.  Histogram normalization addresses this by “standardizing” the lighting conditions across a set of images so that they can be evaluated consistently.

The histogram normalization feature in ADIAT utilizes the match_histograms function in scikit-image.

Input Parameters

1. Reference Image: The image that will be used as the baseline against which all other images will be normalized.

Libraries, Research, and Tutorials

• https://scikit-image.org/docs/stable/api/skimage.exposure.html#skimage.exposure.match_histograms
• https://scikit-image.org/docs/stable/auto_examples/color_exposure/plot_histogram_matching.html
• https://pyimagesearch.com/2021/02/08/histogram-matching-with-opencv-scikit-image-and-python/

Released – October 2023

Synopsis

K-means clustering is a classification technique designed to group like data points.  For the purposes of image analysis, K-means clustering is used to group similar pixels together in order to reduce the overall number of colors represented in an image.  Essentially we are reducing the number of colors in the image so we have “one shade of green”, “one shade of brown”, etc. which allows us to isolate anomalous colors.

The K-Means Clustering feature in ADIAT utilizes the kmeans function in OpenCV.

**Warning** This feature is extremally expensive in terms of performance and execution time.

Input Parameters

1. Number of Clusters: The number of colors we want to remain in the image after processing.

Libraries, Research, and Tutorials

• https://opencv24-python-tutorials.readthedocs.io/en/latest/py_tutorials/py_ml/py_kmeans/py_kmeans_understanding/py_kmeans_understanding.html
• https://opencv24-python-tutorials.readthedocs.io/en/latest/py_tutorials/py_ml/py_kmeans/py_kmeans_opencv/py_kmeans_opencv.html#kmeans-opencv
• https://medium.com/towardssingularity/k-means-clustering-for-image-segmentation-using-opencv-in-python-17178ce3d6f3