Electrocardiogram (ECG) signals are essential for diagnosing cardiovascular diseases, but their accuracy is often affected by noise, including electromyography (EMG). Frequency-based ECG denoising methods struggle to suppress noise in P and T waves without distorting QRS morphology. Stronger smoothing may help but risks losing critical waveform details, indicating a need for adaptive, morphology-preserving approaches. This study proposes a hybrid adaptive filtering approach for ECG denoising that overcomes these limitations, designed to remove EMG noise while preserving the morphology of ECG signal with low complexity design. Two adaptive parallel low-pass filters (LPFs) were used: one for filtering P and T waves (10-25 Hz) and another for filtering the QRS complex (30-70 Hz). The filtered ECG components were then combined to reconstruct a clean waveform, with transition points are smoothed using an adaptive Savitzky-Golay (S-G) filter. The proposed method was tested using SimEMG database (110-pair of ECG recordings from 14-patient with Fs=500Hz). The performance of the proposed method was evaluated by comparing it to related works which used the same database such as adaptive wavelet wiener filtering (AWWF), Wavelet Transform (WT), finite impulse response (FIR), and iterative regenerative method (IRM). For the noisy signals that have SNRin < 4dB, the proposed method achieved an SNR improvement of 10.37dB, outperforming IRM (10.31dB), AWWF (9.23dB), WT (4.59dB), and FIR (4.19dB). For SNRin = 4-8dB, it achieved 10.05dB, surpassing other methods. The method continued to show gradual performance improvements for higher SNR ranges (≥16dB). The average correlation coefficient reached 98.29%, indicating strong signal preservation. The results demonstrate that the proposed method offers a robust, and efficient solution for ECG denoising in real-world biomedical applications.