Learning-Based super-resolution methods usually select several objects with similar features from some examples according to the low-resolution image, then estimate super-resolution result using optimization algorithm. But the result is usually limited by the quality of matching objects and only geometric construction of the images is selected as matching feature, so matching accuracy is relatively low. This paper presents a sparse dictionary model for image super-resolution, which unifies the feature patches of high-resolution (HR) and low-resolution (LR) images for sparse coding. To break through the aforementioned limitations, this method builds a sparse association between HR and LR images, and realized simultaneous matching and optimization methods. The study uses a MCA method to improve the accuracy for feature extraction and carry out super-resolution reconstruction and denoise simultaneously. Sparse K-SVD algorithm is adopted as optimization method to reduce the computation time of sparse coding. Some experiments with real images show that this method outperforms other learning-based super-resolution algorithms.