Background: Filtering chest compression noise from ECG signal is recognized as a promising technique that can help improving cardiopulmonary resuscitation (CPR) quality. Several related approaches have been developed, mainly adaptive filter based on the chest compression property like compression frequency. However, it was found that the existing filtering techniques often lead to significant waveform change, losing vital morphological features in ECG signal and hindering patient assessments in CPR. Methods: Regularization dimension can measure fractal characteristics of one-dimensional signals, which assumed to be different between ECG signal and chest compression noise. We devised a filtering technique named as RD filter using the regularization dimension conception. The RD filter analyzed fractal characteristics of ECG and thoracic impedance signal to determine suitable filter schemes to achieve noise reduction. We collected data from 32 cases of CPR and constructed simulation ECG signals containing compression noise. We compared the performance of RD filter against conventional adaptive filter based on compression rate (CR filter). The filter performance was assessed by signal-to-noise ratio (SNR) and the rhythm morphological similarity between filtered output and actual ECG rhythm was quantified by distance calculation based on dynamic time wrapping (DTW). Results: SNR of RD filter was improved by more than 40% on average compared with CR filter. DTW distance for RD filter was significantly smaller than that for CR filter, indicating the former filtered signal involves more waveform morphology characters regarding ECG rhythm than the latter output. Conclusion: Current results suggest that RD filter can effectively reduce chest compression noise and preserve signal waveform of ECG rhythm better than CR filter. Our proposed approach is a potential valuable technique that can be widely used in CPR to suppress the impact of chest compression and ensure discriminative features of various ECG rhythm.