Deep learning methodologies in assessing logo complexity
Μεθοδολογίες βαθείας μαθησης για την αξιολόγηση της πολυπλοκότητας λογότυπων
In this document, we are going to write about Complexity measure an aesthetic measure and if we can extract features from companies logos that has been rated from some experts. The experiment will use deep Convolutional Networks written in python programming language. The dataset will be 208 images of company logos and we are going to use the original dataset and later we are going to create an augmented dataset from the original images. There are going to be 4 evaluation stage with these 2 datasets in order to conclude about the experiment. In the first stage we are going to use a swallow net with Convolutional Layer, in the second we are going to use a 10 layers Convolutional network, in the third method we are going to use 10 layers but this time batch normalization between the layers. Finally, we are going to use a pre-trained deep network from imagenet and see if it made an improvement in relation to the previous techniques. Furthermore, i am writing about the theory behind the aesthetic measures and the importance in information visualization.