The fixed reference structure is inefficient for video inter coding when handling content with significant motion or scene changes. Moreover, fixed reference structure limits inter coding efficiency since the reference frames exhibiting higher correlation or greater similarity with the current frame may not be selected in a fixed structure. To address the problem, we propose a novel content-aware reference structure optimization scheme. The scheme operates in the pre-encoding stage, leveraging Block-based Image Fingerprints (BIF) to efficiently quantify inter-frame similarity. Based on the BIF metric, the proposed reference structure optimization scheme achieves dynamic I-frame insertion and optimal reference frame selection. Firstly, I-frames are inserted at scene boundaries to prevent prediction error propagation, which is based on a robust scene change detection algorithm that employs an adaptive dual-criterion thresholding mechanism by analyzing both the magnitude and the rate of change of BIF-measured frame similarity. Secondly, for continuous frames in a scene, the algorithm constructs a tailored prediction structure by selecting an optimal set of reference frames based on BIF similarity. Additionally, reference structure is refined using a buffer-aware management module that simulates the buffer constraints of the decoder to prevent buffer overflow at the decoding end. Experimental results demonstrate that our method achieves an average BD-rate of -2.0% while reducing encoding time and maintaining backward compatibility with standard decoders.