Unsupervised Segmentation of Color Images Based on k -means Clustering in the Chromaticity Plane
CBAIVL '99 Proceedings of the IEEE Workshop on Content-Based Access of Image and Video Libraries
Scalable multiresolution color image segmentation
Signal Processing
| Hi-index | 0.00 |
In many digital image processing applications, image segmentation is required to provide initial partitioning of local image regions based on certain statistical or contextual homogeneity measures. One goal of image segmentation would be to segment the image into regions that correspond to physically and semantically coherent objects in the scene. We propose an improved color segmentation algorithm by taking advantage of a simple "k-mode" algorithm and an adaptive Bayesian k-means algorithm. The "k-mode" algorithm uses a physics-based distance metric to generate regular partitioning of the color space. The adaptive k-means algorithm utilizes two additional mechanisms, i.e., spatial homogeneity constraints and spatial adaptivity, to achieve more robust and coherent segmentation. The proposed algorithm integrates a physically more meaningful color space and the corresponding color difference metric into the the adaptive Bayesian K-means framework in an effort towards physics-based segmentation of photographic color images.