This project implemented two distinct approaches for food recognition using a dataset containing 251 classes: a Convolutional Neural Network (CNN) and a traditional method combining Scale-Invariant Feature Transform (SIFT) with Bag of Words (BoW) representation and Support Vector Machines (SVM). Comprehensive preprocessing included image resizing, normalization, and data augmentation. Due to computational constraints, the CNN architecture was limited to under 1 million parameters, resulting in modest performance. In contrast, the SVM with an RBF kernel outperformed the linear kernel variant significantly, though overall accuracy remained relatively low, highlighting challenges in handling high-dimensional image data and computational limitations.