The invention provides a forgery recognition method based on time-frequency analysis, belonging to the field of image processing, comprising the following steps of: decoding a suspicious video streamto obtain a decoding sequence, acquiring an EH function according to the type of a frame in the decoding sequence; determining a distance value in the suspicious video stream, determining the detection frequency based on the first distance value; performing a Fourier transform on the EH function, performing a time-frequency processing based on the S transform on the transformation result to obtaina processing function; determining a detection function based on the processing function, using the detection function on the GOP layer in the suspicious video stream to perform initial positioning,positioning a video stream breakpoint based on two adjacent GOPs that perform the initial positioning, and determining a tampered specific location in the suspicious video stream. Two detection functions are constructed by utilizing the characteristics of all P frames or B frames, and a time-frequency analysis tool S transform is introduced to identify and position the frame tampering operation inthe MPEG stream. The invention reduces the limitations of existing detection techniques and improves accuracy and stability.