This paper proposes a multiframe resolution enhancement algorithm by iteratively estimating and eliminating interpixel interference from neighboring pixels. The rationale is based on the observation that resolution reduction is caused by low-pass filtering nature of the image acquisition device as well as the digital zooming procedure. Both introduce interference between neighboring pixels. At a given location, the interpixel interference is from an integral effect of several low-pass filtering processes, each determined by one of its neighboring pixels. We propose using a Gaussian mixture to model the probability of the integral interpixel interference. The Gaussian mixture is determined by both the local image constraints and the local variation indicator (LVI). Local image constraints come from the image derivative priors, which evaluate the similarity between the current pixel and its neighborhood. A larger image derivative prior implies more interference. We use differences between the current pixel and its neighboring pixels as the image derivative priors. The LVI shows the reliability of the neighboring pixels. The LVI is obtained from both the temporal variation and the spatial variation. Neighboring pixels with larger LVIs are considered less reliable; hence, they provide less information for inferring the interpixel interference. Also, we show from the frequency domain representation that interpixel interference can actually be related to high-frequency loss. By estimating and compensating the missing high-frequency details iteratively, we can recover images with higher resolution. Experimental evaluation on varying inputs, including faces, synthetic text subjects, as well as license plates, validates the algorithm.