In this paper, we propose a new approach that aims to uncover features that can assist in learning celebrity cartoon faces. To recognize cartoon faces, we have tailored the FaceNet architecture. The extracted features are then learned using both a conventional learning model and a convolution model. Furthermore, the Chi-score method is employed to achieve feature reduction to have an efficient yet effective classification. To demonstrate the effectiveness of this approach, we conducted extensive experiments on the Cartoon Faces in the Wild (IIIT-CFW) celebrity cartoon face database, which contains 100 distinct celebrities. In comparison to the existing database, we created 50 different categories namely UOM-Dataset, and examined its performance. The results of the experimentations reveal that the proposed method outperforms several other existing methods, including the state-of-the-art method for celebrity cartoon face recognition.