Blind image super-resolution (SR) is a challenging computer vision task that involves enhancing the quality of low-resolution (LR) images obtained through various degradation operations. Deep neural networks have provided state-of-the-art performance for blind image SR. Prior literature has shown that decoupling the blind image SR task into blurring kernel estimation and high-quality image reconstruction yields superior performance. In this work, we propose a novel optimization problem that leverages geometrical information as prior to estimate blurring kernels accurately. We also introduce a novel blind image SR network that utilizes these estimated blurring kernels within its architecture and learning algorithm to produce high-quality images. Our method employs a curriculum learning strategy, where the SR network training is initially facilitated using ground truth (GT) blurring kernels and later transitioned to the estimated blurring kernels obtained from our optimization problem. Extensive experiments demonstrate the effectiveness of our proposed blind image SR approach compared to state-of-the-art methods across various degradation operations and benchmark datasets.