The state-of-health (SOH) estimation of batteries is crucial for ensuring the safe and stable operation of energy storage systems. However, the methods which use fixed voltage segments as model inputs face limitations in practical applications due to varying user habits, and most of the studies based on incremental capacity (IC) analysis still face challenges in selecting the smoothing parameters. A sliding window smoothing method is proposed to obtain smooth IC curves while simplifying the parameter selection process in real-world applications. This method uses two complementary Wasserstein distance calculation methods to evaluate the differences between the target IC curves and reference curves, which serve as health feature (HF) and are combined with the corresponding starting voltages of voltage segments as model inputs. These inputs are further used to train a Gaussian process regression model. Test results indicate that the voltage segments at any position, even those that do not include the IC curve peak, can be used as model inputs to achieve SOH estimation results with a root mean square error (RMSE) below 2%. In addition, the effects of different smoothing parameters, sampling frequencies and estimation models on the estimation results are also analyzed. The analysis results indicate that the extracted HFs exhibit strong robustness, and the model achieves high estimation accuracy.