Lecture

Image Segmentation

Edge - a rapid change in image intensity or a discontinuity in pixel value.

In edge detection, it requires the use of first and second derivatives.

  1. First Derivative - produces a thick edge profile.
  2. Second Derivative - produces a double-edge profile w/ a sign change

If noises are present in an image, noise must be reduced if edge detection is going to be used.

Edge Detection Algorithms

  1. Image Gradient - measures the change in intensity across the image; derivatves on the x and y directions; also uses derivatives to calculate magnitude and direction of edges within an image.

For the image gradient, we want to find the magnitude and the direction found in an image.

There exist equations to calculate both magnitude and direction:

  • Magnitude Formula - using the Euclidean Distance
  • Direction Formula - using inverse tan ($tan^{-1}$) to calculate the orientation of an edge at a particular point.

Most Famous Edge Detection

The most well-known edge detection algorithm is the Canny Edge Detection Algorithm.

Non-maximal suppression is used to thin out the edges.

NOTE: ZERO-CROSS IS THE SECOND-DERIVATIVE.

Image Thresholding

It simplifies images by splitting them into foreground (object) and background regions based on a threshold value T.

Challenges with image thresholding -

The use of a bimodal histogram is needed/recommended for image thresholding.

Disadvantages of using the Otsu Method:

  1. May not work well with images with more than two distinct regions or non-bimodal histograms.
  2. Applies single glboal threshold–isn’t ideal for images with varying lighting across different areas.
  3. Noise in an image will significantly affect the threshold–will lead to inaccurate segmentation.
  4. Is computationally heavy for high-resolution/multi-level intensity images.
  5. Primarily designed for grayscale images and may not be directly applicable to color images.

Reading

Summary