Image segmentation is the first step in a complex process of object recognition. This report presents a method to gauge the difficulty of segmentation by calculating a scalar parameter Q for an image. This parameter depends on a distribution of the intensity of the grayscale image and the distribution of the clustering of pixels. It is assumed that images with a smaller number of clusters are easier to segment than images with a larger number of clusters. Since segmentation precedes any human perception and categorization, the distribution of parameter Q introduced in this study may be useful in characterizing the variability of images collected in a training dataset for the development of object recognition algorithms which use machine learning (ML) methods. Parameter Q can be especially useful for building a representative dataset of images for training ML algorithms. To demonstrate a link between particular values of Q and different segmentation conditions, a few grayscale images were distorted by some common transformations (such as Gaussian noise, median filtering, and decrease of color depth) and the corresponding values of parameter Q were calculated. To demonstrate a possible use of the parameter Q on data other than grayscale images, depth images of flat planar targets taken by two depth cameras were also processed.
Gauging the difficulty of image segmentation, Technical Note (NIST TN), National Institute of Standards and Technology, Gaithersburg, MD, [online], https://doi.org/10.6028/NIST.TN.2207-upd1, https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=934815
(Accessed December 10, 2023)