An Algorithm for Image Segmentation in HSI Color Space

Abstract

This paper presents a new segmentation algorithm that uses priority components in HSI (Hue, Saturation, Intensity) color space. The algorithm (named SegPC) is designed for HSI color space to support image understanding in different areas of human activities because this color space is modeled to be closer to humans’ color perception. The innovative idea for the usage of all color components with respect to their importance is exploited in algorithm development. Two parameters are used for controlling the segmentation process, the first parameter is the number of colors used in the segmentation process and the second is the order of applying calculated thresholds for color components. The main goal of the research is the analysis of the proposed new segmentation algorithm. A set of images with different types of histograms is used for practical examination of the developed algorithm. The peak signal-to-noise ratio (PSNR) is used as a measure of the effect of image segmentation. The highest PSNR values (≈22 for images with histogram near to bi-modal type and ≈23 for images with histogram with several peak values) are calculated for the priority of color components in which the first analyzed component is color intensity and the number of colors in the segmented image is a quarter of the number of colors in the input image. This result could be explained by the meaning of the intensity component which describes the brightness of the color and because of this, it carries a significant part of color information. The time complexity of SegPC is evaluated through theoretical analysis and experimentally. It is observed that performance time depends on three variables: first, the size of the image strongly influences time complexity (when the size of the image increases then the time for processing increases too), second, the number of colors also is related to time for processing, and third, the selected color components chosen as per the priority has a weak influence on time complexity because it defines the number of pixels which have to be additionally analyzed. The developed algorithm presents promising results. Thus, it could be applied in image analysis and natural sciences.