With the explosive growth of video surveillance data

traditional retrieval methods have become infeasible for practical applications. As an effective alternative

Hashing-based approximate nearest neighbor search has shown great advantages in terms of storage space and time efficiency. However

most existing Hashing methods concentrate on projection learning

and ignore the importance of quantization coding. To address these issues

propose a novel multi-bit quantization encoding method

which constructs new cost function based on Matthews correlation coefficient（ MCC） score and encoding consistency constraint

utilizing genetic optimization algorithm to obtain quantization thresholds. Multiple bits assignment of projection dimension is achieved according to the thresholds. Experimental results on typical dataset verify that the proposed method outperforms existing multi-bit quantization methods.