Magnetic resonance imaging (MRI) images can be used to diagnose brain tumors. Thanks to these images, some methods have so far been proposed in order to distinguish between benign and malignant brain tumors. Many systems attempting to define these tumors are based on tissue analysis methods. However, various factors such as the quality of an MRI device, noisy images and low image resolution may decrease the quality of MRI images. To eliminate these problems, super resolution approaches are preferred as a complementary source for brain tumor images. The proposed method benefits from single image super resolution (SISR) and maximum fuzzy entropy segmentation (MFES) for brain tumor segmentation on an MRI image. Later, pre-trained ResNet architecture, which is a convolutional neural network (CNN) architecture, and support vector machine (SVM) are used to perform feature extraction and classification, respectively. It was observed in experimental studies that SISR displayed a higher performance in terms of brain tumor segmentation. Similarly, it displayed a higher performance in terms of classifying brain tumor regions as well as benign and malignant brain tumors. As a result, the present study indicated that SISR yielded an accuracy rate of 95% in the diagnosis of segmented brain tumors, which exceeds brain tumor segmentation using MFES without SISR by 7.5%.