Arrhythmia is the prime indicator of serious heart issues, and, hence, it is essential to be detected properly for early phase treatment. This article presents an approach for the diagnosis of cardiac disorders via the recognition of 17 types of arrhythmia. The proposed approach includes building a convolution neural network (2D-CNN) which is trained by using images of Electrocardiograph (ECG) signals collected from the MIH-BIH database. The ECGs are first converted into images. This step serves twofold: first, CNN is best suited for classifying image data and thus reduces preprocessing, and second, most ECG recordings are still being produced on thermal paper which can then be captured as image. Next, 2D-CNN is trained and validated. Test results show that the proposed method achieves classification accuracy of 96.67% and error of 0.004%. in addition to the superior accuracy achieved by this method compared to the previous literature, this approach enjoys reduced processing time and complexity apart from the training phase, also by dealing with images this method offers high degree of versatility and can be integrated as utility within other applications or wearables.