In order to solve the practical problem of unsatisfactory restored results, an image super resolution reconstruction algorithm via weighted random forest is proposed. Firstly, features of image patches are clustered by the random forest, and ridge regression is introduced to build the mapping between low and high resolution patches for each type of the leaf note. Then high resolution image patch is obtained by weighted prediction based on the cluster which the test low resolution sample belongs to and the K near neighbor approximate fitting error. Finally, the non-local similarity and IBP(Iterative back projection) are utilized to improve the quality of image reconstruction. Experimental results show that the proposed method effectively improves peak signal to noise ratio and acquires better visual effects in reconstructed image.