Four dynamic thresholding algorithms were implemented and evaluated to segment the images of spray deposits on leaf surfaces. The algorithms were evaluated for accuracy to effectively segment the images that had non-uniform contrast between the droplets and the leaf background, and with varying intensities of the fluorescent tracer over uneven leaf surfaces. The analysis included segmentation of the droplets, estimation of droplet area, and the total number of different size and shape droplet deposits on a leaf surface. Criteria were established to control the morphological operations during the image segmentation process. Guidelines for selecting a suitable dynamic thresholding technique for given image characteristics are proposed. A two-pass approach was evaluated wherein the images were first thresholded by the best general-purpose technique to compute the blob characteristics, based on which an image-specific thresholding algorithm was applied as the second pass for improved accuracy. The experimental results that are provided in this article reveal that this technique is very effective in segmenting images with non-uniform contrasts and could reduce the time required to characterize the spray deposits.
