Cartoons are a stylized visual representation of images that can be utilized to tell stories, convey emotions and express ideas; they are now viewed on the web, educational materials and smartphone applications, leading to greater demand for automated recognition of cartoon characters. This study has taken steps towards fast, accurate cartoon character recognition and categorization, utilizing hand-crafted descriptors only. We developed a dataset of 10 well known characters in randomized poses, styles and backgrounds in order to test robustness. The individual descriptor categories were based on color, texture and shape with separate ensemble classifiers and we subsequently evaluated their scores. As an overview, HSV color-based features had the highest overall accuracy 98.03% from the use of the Random Forest algorithm to recognize our test characters, while the texture based and shape-based features did quite poorly, especially given the abstract nature of cartoon characters. Since Random Forest and Gradient Boosting were the top two scores for accuracy with the HSV features we combined them to see if we could improve our accuracy even further. In our error analysis, we noted that many characters had quite similar colour palettes, causing different characters to look similar enough to be misclassified. The results indicate that we can use hand-crafted features with classical ML classifiers to obtain high accuracy at little computational cost. Practical applications of this work may be in media archiving of large collections and interactive systems. In addition to the technical content, this study also emphasizes recognition and classification system that are simple to implement and inexpensive but have maintained a good level of accuracy and indicates the need for future exploration of hybrid features and large datasets.