Optimized Grayscale Intervals Study of Leaf Image Segmentation

Abstract

In this paper, we propose a segmenting method for leaf images taken in the field. Instead of processing images in the whole grayscale range, we choose several subintervals of grayscale to operate on. The subintervals are chosen in accordance with the envelopes of image histograms. The subinterval is composed of random two local minimum points. In order to examine the effect of the method, four evaluation grades are settled as ‘Perfect (P)’, ‘Well (W)’, ‘Average (A)’ and ‘Unsatisfying (U)’. ‘Perfect’ (P) represents the leaf edge segmented is complete; ‘Well’ (W) represents leaf edge segmented is almost complete; ‘Average’ (A) represents there is no complete leaf edge segmented and ‘Unsatisfied’ (U) indicates that the majority of the leaf edge is not segmented from the background. Operating the segmentation on different subintervals and clustering the segmenting results into these four grades, we could obtain the suitable grayscale subintervals; therefore the subintervals chosen before have been optimized. Experiments on four kinds of leaves including jujube, strawberry, begonia and jasmine leaves shows that optimized grayscale interval solutions are sufficient to segment target leaf from the background with high efficiency and accuracy.