Mango leaf image segmentation on HSV and YCbCr color spaces using Otsu thresholding

Abstract

Research detection of mango tree type that hasn't yet-fruitful needs good result of image segmentation. This is due it use color, texture, and shape as feature. Especially shape feature, we have to produce good image segmentation result as input of feature extraction. For color and texture, we need image segmentation result to be some region of interest in the feature extraction. In this research, we use segmentation by thresholding with Otsu method. We apply Otsu thresholing on Hue, Saturation, Intensity (HSV), and Luminance, Chromaticity Blue, Chromaticity Red (YCbCr) color space for mango leaves. All components of color space are used except Luminance. Segmentation is done by converting input image Red, Green, Blue (RGB) into color space required, then use the color components required, then applying Otsu threshold method, then use several morphology steps to produce good segmentation results. Then the results are compared with ground truth images. Performance testing of color space components provides the best performance component, it is Cr, then Saturation, Cb, Intensity, and Hue respectively. We use Precision, Recall, and F-measure as performance measurement. Precision is a percentage of positive detected in detection result. The Recall is the percentage of real positive detected. While F-measure is weighted harmonic mean of Precision and Recall. The results of empirical testing on components Cr, the average performance of segmentation obtained as follows: Precision is 0.995, Recall is 0.971, and F-measure is 0.983. This performance proves Cr as the right color space component for image segmentation of mango leaves by thresholding.