An image segmentation algorithm for fuel spray schlieren images with noisy backgrounds under engine-like conditions

• Published in: Experiments in fluids 2022-05, Vol.63 (5), Article 89
• Main Authors: Zhou, Yifan, Qi, Wenyuan, Wei, Zhenhong, Zhang, Yuyin
• Format: Article
• Language: English
• Subjects: Algorithms; Diesel engines; Digital imaging; Engineering; Engineering Fluid Dynamics; Engineering Thermodynamics; Fluid- and Aerodynamics; Fuel sprays; Heat and Mass Transfer; High speed photography; Image processing; Image segmentation; Object recognition; Optical measurement; Parameters; Research Article; Signal to noise ratio; Spray characteristics; Statistics; Subtraction; Tags; Texture recognition; Vapor phases
• ISSN: 0723-4864; 1432-1114
• DOI: 10.1007/s00348-022-03435-4

Accurate digital image segmentation is necessary for analyzing the characteristics of fuel spray from various optical measurements. However, some conventional segmentation methods cannot achieve the desired effect for the difficult task with severe noises such as schlieren images under diesel engine-like conditions. In this work, an image processing algorithm named Texture Assessment by Gradient Statistics (TAGS) was proposed for improving the recognition of spray images with noisy backgrounds, which utilizes gradient statistics reflecting texture to distinguish the spray region from the background region. The effects of the key parameters in the proposed algorithm on the segmentation quality were studied. And for validating the applicability and generalization of the proposed algorithm, a series of experiments on diesel evaporating sprays were conducted through high-speed photography of Mie-scattering and double-pass schlieren. It was found that the extraction of gradient statistics can enhance the difference between the spray and the background regions when their gray values are close, which contributes to the image segmentation. For the schlieren images with noisy backgrounds, the signal-to-noise ratio of the TAGS method increased from 1.24 to 9.40 compared with the conventional gray value subtraction thresholding method. The segmentation quality has low dependence on the algorithm parameters of TAGS, and a parameter setting method for its image processing program has been recommended, which enables the analysis of spray characteristics including the macroscopic morphology, the tip penetration and the projected area of liquid/vapor phases. Graphic abstract