The weak light-leaking phenomenon of fibers will definitely decrease the output precision and working reliability of the Interferometric Fiber Optic Gyroscope (IFOG). Unfortunately, because of the small scratches or damages on the surface of fiber cannot be avoided when manufacturing IFOG, it is necessary to develop a method to monitor and evaluate that light leakage phenomenon and harm. In this paper, to evaluate and control that light-leaking defect, on one hand, a lightleaking energy function, which uses the image analysis method and the radiometry theory, is designed to represent the energy loss of light; on the other hand, we build a relationship between that energy function and the assembling quality of IFOG by Support Vector Machine (SVM): (1) we utilize an infrared CCD camera to make a traversal scan of the fibers inside of IFOG. (2) We use graph cut model and flood fill algorithm to make a segmentation of that light-leaking region. (3) We define a new energy function to estimate the light leakage level. (4) We use SVM to build a connection between the light-leaking energy function and the final output of IFOG. Here the image energy function is regarded as the training data of SVM, while the bias of IFOG is treated as the supervising data of it. A practical application case of IFOG light path quality evaluation is shown in the end of this paper.