Abstract: To improve the Bidirectional Motion Estimation (BME) accuracy, the Context Cube (Con-Cube) matching is proposed, which uses the compact context features to suppress the block mismatches. Firstly, for the patch centered at each pixel in a block, its correlation surface is produced by using the self-similarity descriptor, and the histogram of the correlation surface is extracted as the context feature of pixel. Secondly, the Con-Cube is constructed by combining the context features of all pixels into the block along a new dimension. Then, based on Con-Cube, the bidirectional context matching is designed to generate the motion vector field of the interpolated frame. Finally, the interpolated frame is produced by linear interpolation compensation. Experimental results show that the proposed algorithm can improve both objective and subjective quality of the interpolated frame when compared with some traditional algorithms, indicating that Con-Cube can effectively prevent the block mismatches existing in BME.