Related Experiment Videos
An automatic color segmentation algorithm with application to identification of skin tumor borders
S E Umbaugh1, R H Moss, W V Stoecker
1Electrical Engineering Department, Southern Illinois University, Edwardsville 62026.
Summary
This study introduces a principal components transform algorithm for automatic color segmentation in skin tumor images. The method accurately identifies tumor borders across various color spaces, outperforming traditional methods.
Area of Science:
- Medical imaging
- Computer vision
- Image processing
Background:
- Accurate tumor border detection is crucial for effective skin cancer treatment.
- Traditional segmentation methods can be limited by color variations and image complexity.
Purpose of the Study:
- To develop and evaluate a principal components transform algorithm for automatic color segmentation of skin tumor images.
- To compare the performance of this algorithm across six different color spaces.
Main Methods:
- A principal components transform algorithm was applied to 500 digitized skin tumor images.
- Images were analyzed in RGB, IHS, spherical, chromaticity, CIE, and CIE-LUV color spaces.
- Automatic induction determined the optimal number of colors for segmentation.
- Performance was compared against a luminance radial search method.
Main Results:
- The principal components transform captured 91% of image variance along the principal axis.
- The algorithm performed comparably or 10% better than the luminance radial search method in border detection.
- The spherical transform yielded the highest success rate, while chromaticity had the lowest error rate.
Conclusions:
- Principal components transform offers a robust method for automatic color segmentation in medical imaging.
- The choice of color space significantly impacts segmentation accuracy, with spherical and chromaticity transforms showing distinct advantages.
- Further research is needed to address data variances for definitive statistical comparisons.