Abstract: Aiming at the problem of over-segmentation and under-segmentation in tea sprouts segmentation under natural environment, a new method based on sharpness function evaluation and color clustering was proposed. The image regions with high sharpness function were obtained by combining Tenengrad gradient evaluation and slider analysis, then K-means clustering segmentation under RGB, HSV, Lab, YCbCr color models were finished. Experiments showed that the upper quartile of Tenengrad gradient value as the threshold, the missed selection rate was 25%;Under the HSV color model, the segmentation of tea sprouts was achieved by K-means clustering methods, and the segmentation accuracy in sunny and cloudy were 72.48% and 77.83%. Compared with the direct K-means segmentation method, the false positive rate are reduced 5.19% and 2.03%. The method could segment tea sprouts in natural environment effectively,reduce under-segmentation accuracy and over-segmentation rate, provide theoretical reference for intelligent picking.

Key words: tea sprouts, sharpness function, cluster, image segmentation