Anal Quant Cytol Histol. 1996 Aug;18(4):323-9.
Analytical and quantitative cytology and histology
J Smolle
PMID: 8862676
OBJECTIVE: To present an iterative algorithm that selects the optimal pair of grey level image components of red-green-blue (RGB) and hue-luminance-saturation (HLS) color image models and the optimal weighting factor for automated segmentation of the structure of interest in color images.
STUDY DESIGN: The method was based on a stepwise technique, searching for the image combination that yielded the maximum two-sample t value for the grey levels of the object of interest and of the background. As a result, a transformed grey level image providing a two-peak grey level histogram suitable for automated thresholding was provided.
RESULTS: The segmentation of collagen bundles in hematoxylin and eosin-stained slides of cutaneous melanomas is presented as an example. Though each of the RGB and HLS grey level images showed extensive overlap of objects and background, the combined image facilitated clear-cut discrimination.
CONCLUSION: The proposed algorithm permits the discrimination of specified objects in color images that are not characterized by grey level in any of the underlying grey level image components.