This paper presents a new SD-OCT outer retinal boundary identification method based on the improved graph-theoretic approach in SD-OCT retinal image, which is robust to the image quality degradation and the pathological morphology variability. The performance of the proposed method was verified using the SD-OCT image database with inherit retinal dystrophies, which suffer from the artifacts most among different macular degeneration diseases. The experimental results of the subjective evaluation indicated that the identification results using the proposed method was substantially improved compared with the current built-in software in the SD-OCT devices.