Real-time automatic multilevel color video thresholding using a novel class-variance criterion

摘要: Color image segmentation is a crucial preliminary task in robotic vision systems. This paper presents novel automatic multilevel color thresholding algorithm to address this efficiently. The proposed consists of learning process and multi-threshold searching process. learns the distribution an input video sequence HSV space, automatically determines optimal multiple thresholds segment all colors-of-interest based on class-variance criterion. For process, simple efficient color-distribution operating with color-pixel extraction method learn model images, which simplifies search for through conventional method. nonparametric extended within-class variance criterion find upper bound lower threshold values each channel. Experimental results validate performance computational efficiency by comparing three existing methods, both visually quantitatively.