Photo aesthetics evaluation system: an application of CNN and SVM

In this project, we applied two machine learning techniques: CNN (Convolutional Neural Network) and SVM (Support Vector machine) to build an image aesthetic evaluating system. And we have achieved an 5-folder cross validation accuracy of above 99% by using CNN implemented in Torch. In the ‘Introduction’ section, a brief background of the problem and an introduction of our system are given. In the ‘Dataset and Features’ section, we’ve discussed how we generate the training data and extract the input features for the machine learning algorithms. And in the next two sections: ‘Methods’ and ‘Experimental Result’, how we applied CNN and SVM and the performance of each algorithm have been discussed. The report ended up with a ‘Conclusion’ section including the summary of work and a list of future work. Keywords—image classification, SVM, CNN, machine learning