The piecewise statistical stationary property of natural image signals provides an effective way to model the image signals. However, the statistical stationary region always has an irregular profile. Large mismatch errors rise when a regular shape window is adopted to estimate the statistics in these regions. This paper proposes an implicit piecewise autoregressive model based on a probabilistic description of the statistical stationary region. According to this model, an improved image interpolation algorithm is presented. The experimental results show that the proposed interpolation method can greatly alleviate the blurring, ringing, and noisy artifacts around edges in interpolated image.