The invention discloses an image super-resolution reconstruction method based on a local regression model. The method comprises the following steps: at first, carrying out Gaussian low pass filtering on an input low resolution image to obtain a low frequency band image thereof, carrying out bicubic interpolation to obtain an approximate low frequency band image of a high resolution image; then, applying a one-order regression model to each image block in the low frequency band image of the high resolution image during reconstruction, wherein a mapping function between high / low images in the regression model can be obtained by a machine learning method of an input image, namely, sampling corresponding positions of the input low resolution image and the low frequency band image thereof to obtain sampling image blocks of corresponding positions, and carrying out dictionary training; and finally, respectively applying the one-order regression model to non-local self-similar blocks of the reconstructed image blocks, and carrying out weighted integration to obtain reconstructed high resolution image blocks. By adopting the method provided by the invention, no external image model is required, a prior model is obtained by learning the input image, and the high resolution image reconstructed by the model has better subjective and objective reconstruction effects.