A Method to Increase Class Separation in the HS Plane for Color Segmentation Applications

In this paper, an alternative to increase the separation between two classes (object and background) projected in the HS color plane (HSI space) is presented. Our proposal is a preprocessing stage that improves the segmentation process of color image sequence frames in real-time. The authors propose to perform a linear transformation to each frame in the RGB space that consists on adding a vector in RGB components (color vector). The optimal color vector to add maximizes the class separation in the HS plane, improving the segmentation process. The improvement in the class separation is achieved for the non-linearity of the transformation between the RGB and the HSI spaces. The C₁C₂ chromatic sub-space (YC₁C₂ space) is used to obtain this optimal vector to add. Its effectiveness has been tested especially in images with a reduced contrast between the colors of the object and background and when the size of the object to segment is very small in comparison with the size of the captured scene.