Image Classification using SVM and CNN

On the surface, teaching a computer to do something like image classification seemed very intriguing to us. Moreover, there are countless real-world applications of this concept. It is in light of these reasons that we decided to work on Image Classification. Thankfully though, this topic has been well-researched by the scientific community and we didn’t break a sweat finding resources to learn from. So naturally, we perused a bunch of research papers that dealt with image classification, each from a different perspective. We then decided to implement image classification on a small-scale with the limited hardware we were in possession of. As difficult as it was, we started with SVM and a very small dataset to achieve an accuracy of 93%. Although SVM is a very strong technique, achieving such a high accuracy is still an anomaly. We realized that our results boasted such a high accuracy due to the lack of a large enough dataset. So, using data augmentation, we more than tripled the size of our dataset. On performing SVM now, we achieved an accuracy of 82%, a significant decrease. Unsatisfied with the results, we decided to move to other deep learning techniques. This quest led us to Neural Networks and, CNN. On successfully implementing CNN, we achieved an accuracy of a staggering 93.57% on the very same dataset. This stands as a testimony to the increased potential of deep learning techniques over the more traditional machine learning techniques.