This paper introduces a new local polynomial regression (LPR)-based high-resolution image reconstruction method for human brain magnetic resonance images. In LPR, the image pixels are modeled locally by a polynomial using least-squares (LS) criterion with a kernel having a certain bandwidth matrix. Steering kernels with local orientation are used in LPR to adapt better to local characteristics of images. Furthermore, a refined intersection of confidence intervals (RICI) adaptive scale selector is adopted to select the scale of the steering kernels. The resulting steering-kernel-based LPR with RICI (SK-LPR-RICI) method is applied to reconstruct a high-resolution brain MRI image from a set of low-resolution MRI images. Simulation results show that the proposed SK-LPR-RICI method can effectively improve the image resolution and peak signal-to-noise ratio.