Use of thresholding algorithms in the processing of raindrop imagery

Abstract

Several thresholding algorithms are applied to the analysis of drop images, and their performance is compared. Images were obtained by use of a digital camera setup in which drops were illuminated from behind, resulting in an image of the drop silhouette. Each algorithm was evaluated based on the accuracy of the drop diameter obtained from the thresholded image and on the size of the depth of field. Because of the difficulty associated with creating drops that have a known diameter, solid spheres composed of a glass with an index of refraction close to that of water were used in computing the depth of field and in determining the accuracy of measured diameter. The application of this study is to the automatic measurement of raindrops and images were obtained during several storms. With each thresholding algorithm this raindrop imagery was used to compute the probability density function of drop diameter, and the rain rate. The performance of each thresholding algorithm was quantified by comparison of these measurements with simultaneous measurements obtained by use of a Joss-Waldvogel disdrometer.

Edge detection methods applied to the analysis of spherical raindrop images

“…Saxena and Saylor [34] utilized a boundary counting algorithm for this purpose. Because the characteristics of the holes observed in the images generated by edge detection algorithms are different than those obtained using the thresholding methods presented in Saxena and Saylor [34], some modifications of their boundary counting algorithm would be needed. Additionally, that algorithm did not work perfectly, occasionally classifying out-of-focus drops as in-focus.…”

“…Hence, this present work does not demonstrate the ability of this system to image drops at terminal velocity. However, as noted earlier in this paper, Saxena and Saylor [34] employed the setup used in the present work to obtain raindrop images without smearing. The setup shown in Fig.…”

“…Image segmentation processes are broadly classified into the following three groups [32,33]: a) thresholding or clustering techniques; b) region growing and c) edge detection techniques. Thresholding methods have been investigated in our earlier work and will not be discussed here [34,35]. Region growing techniques have a high degree of complexity associated with them and were not considered to be practical for this application.…”

“…In this situation, the problem of avoiding image smearing ultimately reduces to one of providing sufficient lighting intensity. In the work of Saxena and Saylor [34], the setup used in the present work was used to obtain DSDs using imagery acquired during an hour of rain, and using a 300 W halogen lamp. Smearing was not observed for the raindrop images acquired in that work.…”

Application of a histogram modification algorithm to the processing of raindrop images

“…Further details of the experimental setup can be found in Saxena and Saylor. 13 Images were acquired over a 30-cm range of z, centered on the focal point. The images were acquired far from the camera and then progressively inward, crossing the focal point and then continuing closer to the camera.…”