Fuzzy inference system for region segmentation using the YCbCr color model

Segmentation is one of the main tasks in image processing and pattern recognition systems. In this paper, a segmentation technique of color images based on fuzzy inference model is proposed. Triangular membership functions are used in the input of the fuzzy system, the Mamdani type fuzzy inference system is applied and for the output universe, singleton-type functions are used. To get the accurate output value, the weighted average method is applied. The YCbCr color space is used as feature space. The fuzzy membership functions characterize the different membership levels between hue and Chroma from the YCbCr color model. The fuzzy inference system classifies data and generates regions of pixels with an homogeneous color level in the output images. The proposed technique was also applied to the RGB color space and the results were compared; the best results were obtained in the YCbCr color space. In this color model, the changes of hue in presence of illumination variations are considered, so that it has a better performance in the segmentation task; in addition, the processing time was lower in this color space.