Background Segmentation using a Gaussian Mixture Models & VGG16/UNet Deep Learning model. Corner detection additionally implemented for the deep learning model.
GMM model implemented from first principles. The GMM attempts to cluster hand-crafted features to perform segmentation of the input images as either background or foreground.
90+ % accuracy
VGG16/Unet (Keras module). The famous UNet architecture constructed with the VGG16 pre-trained ImageNet weights. Includes a third class of corners, to detect the corner locations of the puzzle images.
99+ % accuracy