Abstract: The scratches on fiber optic imaging elements are predominantly detected manually, resulting in highly subjective inspection outcomes. To eliminate the influence of human subjectivity in detection, this study employs a platform equipped with multi-angle annular top lighting and backlighting to capture surface images of fiber optic imaging elements. A hybrid strategy-based image recognition algorithm is proposed. The algorithm first defines the image processing region using backlit images, filters out top-lit images containing large bright spots via grayscale histogram analysis, then detects scratches using an adaptive dual-threshold edge detection algorithm and a differential edge extraction method. The combined results from both processes undergo connected domain analysis to identify scratches. Experimental results indicate that the proposed algorithm achieves a scratch detection rate of 93%, while enabling rapid and effective detection of faint scratches.