Noise image segmentation method based on offset field estimation

摘要

The model of level set based on the local offset field solves the problem that the traditional local model cannot deal with the uneven grayscale images. But the performance of the model on noisy images is still not excellent. In response to this problem, this paper introduces a convolutional autoencoder to design an unsupervised noise separation mechanism to model the noise field based on the local model, so that the additive noise field can be separated from the image and avoid its interference to the image segmentation process. The results show that the proposed method can separate the additive image noise effectively and improve the segmentation accuracy in a noisy environment, which is superior to traditional noise robust models and offset field models.