An Image Segmentation Based on a Piecewise Smooth Model in Deep Learning

We propose simple method that can solve the image segmentation problem in unsupervised manner. Our convolutional neural network is based on ResNet-18 to construct auto-encoder structure since the results with the ResNet-18 are reasonable. Our network optimizes the traditional segmentation algorithms: Chan-Vese algorithm and Mumford-Shah algorithm. Compared to some unsupervised segmentation algorithms that need additional modules to cover the inaccurate boundaries, our method can be trained end-to-end and consists of a simple objective function.