Segmentation of Chronic Wound Areas by Clustering Techniques Using Selected Color Space

This work describes the segmentation of wound areas for chronic wound assessment through appropriate color space selection. For this purpose, wound images grabbed by normal digital camera were preprocessed using combined gray world and retinex method for color correction. Thereafter a 5 x 5 median filtering and anisotropic diffusion were used for noise reduction and color homogenization respectively. Here fifteen color spaces were compared based on mean contrast between wound and non-wound regions where D-r and D-b chrominance channels of YDbDr color space were found to provide the highest contrast. Then the wound regions were segmented using clustering techniques viz., k-means and fuzzy c-means on D-r and D-b color channels. The wound segmentation accuracies were compared which showed 74.39% (k-means) and 72.55% (fuzzy c-means) accuracies in D-r Channel whereas 73.76% (k-means) and 75.23% (fuzzy c-means) in D-b channel. Overall, fuzzy c-means algorithm in D-b channel provided higher accuracy for wound segmentation.