Chapter 10 Clustering (CP467) Flashcards

Image Processing & Pattern Recognition Image Segmentation and Clustering Part 2: Clustering

clustering

grouping together similar points and representing them with a single token

uses and application for clustering

uses: for tasks like segmentation in image processing (grouping pixels based on intensity, color, texture)

applications: object detection, image segmentation and compression

criteria for forming clusters

• intensity (grouping pixels w similar brightness or grayscale values)

• color ( grouping pixels w similar color values )

• texture (grouping pixels based on patterns or surface characteristics)

goal for k-means clustering

partition data into k clusters by finding k representative centers.

k-means clustering steps

• initialize k clusters at random

• assign data point to nearest cluster center

• update each cluster center w the mean of the points assigned to it

• repeat 2nd and 3rd steps till cluster centers stabilize

goal of clustering in image segmentation

group pixels into clusters based on similarity in intensity, color, texture, or other features

how does k-means clustering segment an image

by grouping pixels into k-clusters where each pixel is assigned to the nearest cluster center based on intensity or other features

impact of choosing k in k-means clustering

a large k produces more detailed segmentation, while a smaller k creates coarser segmentation

feature space for clustering based on intensity

1-dimensional space where each pixel is represented by its intensity value

what is the feature space for clustering based on color similarity

a 3-dimensional space representing RGB (or other color channels) where each pixel’s coordinates are its color values

how does clustering based on texture similarity work?

it uses filter responses to represent each pixel in a high-dimensional feature space

why might intensity alone not be sufficient for segmentation

pixels w the same intensity but different locations (e.g. two black areas) may belong to different segments.

how does combining intensity and position improve segmentation

by adding x, y coordinates to the feature space, clusters can be formed based on spatial proximity as well as intensity

pros of k means clustering

• simple and fast

• easy to implement

• converges to local minimum of within cluster squared error

cons of k means clustering

• specifying k

• sensitive to initial centers

• sensitive to outliers

• detects spherical clusters only

mean shift clustering

clustering technique that improves upon K means by:

• not requiring k in advance

• handling arbitrary cluster shapes

• being robust to initialization

what are the main limitations of K means that mean shift addresses

• k means requires the number of clusters(k) in advance

• k means is sensitive to initialization

• k means cannot handle non-spherical clusters effectively

key steps in the mean shift clustering algorithm

• center a window on a data point

• identify all points within the window and compute their mean

• shift the window to the mean location

• repeat steps 2-3 until convergence

what happens after convergence in mean shift clustering

• each centroid represents a cluster

• assign all data points to the nearest centroid to finalize the clusters

• merge centroids that converge very close to each other (optionally)

how does mean shift define cluster shape and size

by the window size(h):

• a larger window results in fewer cluster

• a smaller window captures finer details and more clusters

key strength of mean shift

versatile and can handle clusters of arbitrary shapes and sizes without needing to specify the number of clusters(k)

mean shift strengths and weaknesses