Mean-shift-based color segmentation of images containing green vegetation

Separating green vegetation in color images is a complex task especially when there are noises and shadows in the images. Our objective is to improve the segmentation rate of the images containing green vegetation by introducing a mean-shift procedure into the segmentation algorithm. The proposed algorithm mainly consists of two stages—feature extraction and image segmentation. At the first step, multiple color features, such as hue and saturation in HSI color space were extracted, as well as red, green and blue value in RGB color space. At the second step, with the extracted features, mean-shift segmentation algorithm and a BPNN, the image was classified into two parts: green and non-green vegetation. The algorithm’s performance was assessed on 100 images, which were acquired under field conditions, covering different plant types, illuminations, and soil types. The test showed that the median of mis-segmentation of green and non-green vegetation of proposed method is about 4.2%.