Research On Plant Leaf Segmentation Algorithm In Complex Background

The automated identification and classification of plants play a very important role in plant-related applications.As the leaves of different plants in the shape and texture have a big difference,the classification of plants can be done by the leaves.Leaf segmentation is the basis for the identification and classification of plant leaves.The degree of leaf segmentation will directly affect the accuracy and precision of leaf identification and classification.Therefore,how to use digital image processing technology and machine vision technology to segment the leaf image is particularly important.However,plant leaves are usually in a complex natural background,subject to different lighting,leaf overlap and other factors,making the leaf segmentation still a more difficult problem to solve.Therefore,there are more and more researchers interested in this issue.In this paper,the algorithm of plant leaf segmentation in complex background is studied on the existing image segmentation algorithm.A segmentation algorithm based on K-means clustering and adaptive threshold analysis,and a marker-controlled watershed segmentation algorithm based on morphological reconstruction are proposed.The main research work and innovation of the paper are as follows:(1)A plant leaf segmentation algorithm based on K-means clustering and adaptive threshold is proposed.Firstly,this method defines a K-means clustering method for selecting the optimal clustering number in CIE L*a*b*color space,only clustering the color components(a*and b*)and reducing the impact of the luminance in the images,it can roughly separate the object and background information.Then,an adaptive threshold segmentation method based on multi-technique fusion is defined to further segment the leaves.Based on Sobel filtering,histogram,connectivity and image entropy techniques,the optimal segmentation threshold is selected,and applied accurate segmentation among the overlapping parts of the leaves.Lastly,the mathematical morphology is used to refine the segmentation results,fill the holes in the object and smooth the edges.The experimental results show that the algorithm can effectively segment the object from the complex background,and it can also remove the background information and keep the object remain complete.(2)This paper proposes a marker-controlled watershed segmentation algorithm based on morphological reconstruction to realize the effective segmentation of plant leaf images in complex background.The algorithm scales the image size proportionally,filters the scaled leaf image and obtains the gradient image.In the same time,the scaled leaf image is processed by morphological reconstruction to find the best-marked image for the original gradient image.The method can reduce the number of regions of plant leaf images in complex background and solve the over-segmentation problem in the watershed segmentation effectively.Then,the modified gradient image is segmented by a watershed transform,which is different from the traditional method of marking after the morphological reconstruction.The method only needs scale image,and it effectively solves the adjustment of the sizes of multiple structural elements during the morphological reconstruction and marking.The segmentation result is further refined by threshold segmentation and mathematical morphology for the image with inaccurate edge segmentation of the object,so that the complete object is preserved while the background is removed.The experimental results show that the algorithm can effectively solve the problem of over-segmentation in watershed segmentation and extract the object accurately from complex background.