Interactive Image Segmentation Using Multiple Color Spaces

Image segmentation is an important first task of any image analysis process. All the subsequent tasks, such as feature extraction and object recognition, rely heavily on the quality of the segmentation. This paper presents the perceptual color space and its advantage in segmenting the color image interactively. Although the color space reflects the way man observes color, its strangeness is a problem not to be overlook. A small change of color value can cause hue component (H) to fluctuate largely when saturation parameter (S) of HSL color space is less than 10% or larger than 90% or the three components' values of RGB color space are close to each other. Both RGB and HSL color spaces are used in image segmenting in order to avoid the strangeness. The former is used if color distribution is in the range of the strangeness, otherwise the latter is used. A region growing algorithm and a merging algorithm using multiple color spaces and multiple homogeneity criterions are proposed. The two algorithms take into account color similarity and spatial proximity simultaneously. The homogeneity criterions are relative to seed point, expand point, and growing region respectively. The history of Dunhuang's frescoes has nearly two thousand years. The originals, owing to discoloring and partly dropping for various reasons, had been damaged seriously. The mural image is more difficult to process than other types of image. Fully automatic general segmentation is an unsolved problem in the complex applications.The interactive image segmentation technique has been used in the image processing of DunHuang mural and has achieved effective result.