Abstract: Objective and Methods Normal moveout (NMO) correction is a core step in seismic data processing. However, conventional methods often cause wavelet stretching when processing wide-angle or long-offset seismic data, leading to degraded stacking quality and distorted amplitude variation with offset (AVO) analysis. While matching pursuit NMO (MP-NMO) can mitigate stretching effects, it suffers from lateral continuity due to overlapping seismic events. To address this limitation, this study proposes an improved MP-NMO method based on multi-parameters fitting. During matching pursuit decomposition, cubic polynomial fitting is applied to the amplitude, frequency, and phase of wavelets from the same reflector across different offsets, eliminating local anomalies. A second-order time-distance curve is used as the initial value for adaptive time scanning within a dynamic window to determine optimal arrival times, avoiding the complexity of higher-order traveltime equations. Additionally, multi-scale Morlet complex wavelet scanning enhances noise robustness and frequency adaptability. Results Theoretical tests demonstrated that the proposed method significantly reduces wavelet stretching at long offsets with amplitude and frequency errors below 5%, which is more than 90% higher than traditional matching pursuit NMO correction (error up to the 40% ). Noisy model test indicates this method exhibits robustness against noise and the convergence can be achieved with 7 iterations. Field data applications show that enhanced continuity in shallow long offset gathers and improved resolution in stack sections, making the events between 0.8 to 1.7 seconds more clearer. Conclusions This method provides a high-fidelity solution for seismic data processing in complex structural areas, particularly benefitting AVO/AVF attribute analysis and anisotropic inversion.