PuzzleNN: A Neural Network for Image Segmentation Based on Clustering

In order to create a system for Image Semantic Segmentation that works similarly to the human way of approaching visual problems, we decided to use divide-and-conquer to break an image into smaller pieces, called microimages. We use Spectral Clustering to understand the features and group them to identify the location of different objects in the image. We trained a Convolutional Neural Network to compute the affinity matrix required by Spectral Algorithm. We managed to obtain competitive results on the Berkeley Segmentation Dataset and we hope to continue to improve the method.