ABSTRACT: This paper presents a hybrid and unsupervised approach to flame front detection for low signal-to-noise planar laser-induced fluorescence (PLIF) images. The algorithm combines segmentation and edge detection techniques to achieve low-cost and accurate flame front detection in the presence of noise and variability in the flame structure. The method first uses an adaptive contrast enhancement scheme to improve the quality of the image prior to segmentation. The general shape of the flame front is then highlighted using segmentation, while the edge detection method is used to refine the results and highlight the flame front more accurately. The performance of the algorithm is tested on a dataset of high-speed PLIF images and is shown to achieve high accuracy in finely wrinkled turbulent hydrogen-enriched flames with order of magnitude improvements in computation speed. This new algorithm has potential applications in the experimental study of turbulent flames subject to intense wrinkling and low signal-to-noise ratios. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s00348-023-03651-6.