Incremental capacity analysis is a popular tool for the evaluation of state-of-health in battery management. In digital systems, the incremental capacity is generally approximated with the ratio of the capacity difference to voltage difference (ΔQ∕ΔV), which unavoidably amplifies measurement noises. To enhance its resilience against noises and improve the estimation accuracy, a two-dimensional filter is designed by employing historical information from both time and batch (cycle) directions inspired by batch-wise repetitiveness of the incremental capacity trajectories. Specifically, in the batch direction, a Luenberger observer is utilised to provide a batch-to-batch smoothing at the beginning of each charging cycle, while in the time direction, a bias-corrected Gaussian moving average filter is applied to smooth the incremental capacity value with respect to the voltage at every sampling time. Experimental results show that the root-mean-square-error of the proposed filter is 50% lower than the benchmark algorithms, and the noise sensitivity is significantly reduced by 93%. When using incremental capacity peaks extracted from the proposed filter for state-of-health modelling, the width of the 99% confidence interval would be narrowed by 45%. Moreover, the model-free nature of the proposed method enables its application to different batteries, paving a reliable way for effective battery health assessment. •The incremental capacity values are modelled as a function of voltage.•Bias-corrected Gaussian-moving-average filter is developed for real-time filtering.•Symmetric windows could be used for filtering without violating the causality.•Batch-wise Luenberger observer is developed for curve smoothing in each cycle.