Weld interface defects at a one-micrometer level can be identified with the support of the optical microscope (OM) images. However, identification of such defects below one-micrometer becomes challenging due to the limited resolution of OM imaging. Besides, the diagnosis gets even more challenging when the OM equipment is in worn-out condition. This research develops a hybrid sparsity model for efficiently improving the resolution of the degraded weld microstructure images. Rather than using a colossal patch databank for training, we deploy two condensed well-learned dictionaries. The sparse information is recovered by deploying a low-resolution (LR) and high-resolution (HR) dictionary; with the support of these dictionaries the resultant high-resolution weld microstructure image is computed. We optimize the hybrid sparsity algorithm utilizing the Sparsity-Enabled Single Value Decomposition (SE-SVD) algorithm. The calculated weld microstructure image dynamically chooses the suitable patches from the dictionary for achieving the most elegant representation among all patches for the available LR weld microstructure image. The proposed approach is resilient to noise, as it accomplishes the tasks of noise-removal and resolution enhancement at the same time. The experimental results indicate that the proposed model surpasses certain peers concerning the algorithm speed, effectiveness, and overall performance, aiding in better diagnosis of weld interface defects. © 2020 Taiwan Academic Network Management Committee. All rights reserved.