Self-Adaptive Superpixels Based on Neural Network Models

In this paper, the self-adaptive superpixels are generated based on a neural network model. Superpixels are clusters of pixels, which can simplify the expression of images. Superpixels are widely used in the field of video/image processing. However, existing algorithms are mainly based on hand-crafted features, which will lose the details of the images. We use the neural network model to extract the deep features of the pixels instead of the hand-crafted features. A predicted object area is obtained according to the results of the neural network models. Self-adaptive superpixels are generated by the clustering algorithm based on the deep features of the pixels and the predicted object area. Finer superpixels are generated in the object area, and coarser superpixels are generated in background area. The generated self-adaptive superpixels can represent the image in a concise way and improve the segmentation accuracy. Experimental results show that the proposed algorithm outperforms several state-of-the-art methods on the BSDS500 dataset.