The GrabCut Algorithm for the Automatic Segmentation of Target Leaves under the Complex Background

Since the traditional GrabCut algorithm needs to manually select the target area and it is also difficult to accurately segment the target leaves from their image with the complex background or the uneven illumination, an automatic segmentation algorithm based on the GrabCut algorithm is proposed. Combining with the fuzzy Gaussian mixture model (FGMM) and the leaf color information, this algorithm can automatically segment the target leaves from their image with the complex background or the uneven illumination. Firstly, using the appropriate fuzzy factor, the image pixels are labeled by the fuzzy Gaussian mixture model and it is used as a first-labeled image. The second-labeled image which is obtained by the Extra-green algorithm (EXG) with the appropriate threshold is used to initialize the GrabCut algorithm and the segmentation of the target leaves is accomplished automatically. The validity and the misclassification rate of the algorithm are investigtaed using several different samples. The results show that this algorithm can automatically segment the target leaves under the complex natural background or the uneven illumination and the average error rate is 1.625.