Optic fiber imaging elements are used in weak visible light, X-ray imaging and high-energy particle detection imaging devices. They play an important role as input and output window materials of image intensifiers. Optic fiber imaging elements are arrays of tens of millions of micron-scale single optical fibers arranged regularly. The fabrication process requires several times of fiber drawing and secondary thermal processing such as hot melting pressure, torsion, and stretching. After these processes, there may be spots and linear chicken filaments that are called defects existed on the interface among the fibers and multi-fibers. Due to these defects, the quality of imaging is seriously reduced, and even the misjudgment or omission of image signal recognition can are caused. How to detect such defects has no an ideal solution. Currently, non-quantitative microscopic observation is generally used. This method, however, is high in misjudgment and low in detection efficiency. In this paper, a device for automatic detection of optical fiber image defects based on machine vision algorithms, including its working principle, structure, detection steps and characteristics are introduced. The device not only can automatically measure the size of each defect, but also can count the defect distribution according to the quality zones. The test results are stable and accurate. It is especially suitable for batch detection and research of optic fiber imaging elements.