Lip shape counterfeit video detection method and system based on multi-phoneme selection

The invention relates to a lip shape counterfeit video detection method and system based on multi-phoneme selection, and the method comprises the steps: S1, building the mapping of phonemes and lip shapes, and obtaining a phoneme category set; s2, selecting a corresponding multi-phoneme category subset from phoneme categories according to different lip shape tampering methods, and constructing at least one phoneme lip shape data set as a training sample set according to a preset lip shape frame number; s3, according to the at least one phoneme lip shape data set, constructing at least one corresponding lip shape forgery video detection sub-model, and respectively performing training to obtain at least one corresponding lip shape detection result; and S4, fusing the at least one lip shape detection result to obtain a fused lip shape detection result. According to the method, a phoneme-lip shape mapping relation is established, phoneme-lip shape subset training sub-models are selected, and results are fused. Ac