Cloud computing has revolutionized the access to and the use of emerging computing hardware technologies. To achieve efficient resource management and provisioning in cloud environments, we need mechanisms to predict future resource utilization. Cloud resource utilization traces of modern applications and use cases are rather complex and random, rendering traditional time series forecasting methods ineffective. Recently, several machine learning methods are being employed to build more sophisticated prediction models, but fail to provide highly accurate predictions especially for longer windows of time in the future. Recent work in the financial domain shows how the use of image representations of time series data, and relevant image-based methods can lead to more robust and effective forecasting. To this end, we explore the use of visualization, computer vision and image-based machine learning methods to forecast future resource utilization in cloud environments. Our analysis shows that the proper visualization of the raw data coupled with image-based machine learning methods similar to those used in video frame prediction, is able to accurately forecast cloud resource utilization for long windows of time in the future, and produce robust forecasts of unseen data that exhibit very different trends from the training dataset.