Superpixel segmentation with squeeze-and-excitation networks

• Published in: Signal, image and video processing Vol. 16; no. 5; pp. 1161 - 1168
• Main Authors: Wang, Jingjing, Luan, Zhenye, Yu, Zishu, Ren, Jinwen, Gao, Jun, Yuan, Kejiang, Xu, Huaqiang
• Format: Journal Article
• Language: English
• Published: London Springer London 01.07.2022; Springer Nature B.V
• Subjects: Algorithms; Artificial neural networks; Computer Imaging; Computer Science; Downstream effects; Excitation; Feature maps; Image processing; Image Processing and Computer Vision; Image segmentation; Machine learning; Modules; Multimedia Information Systems; Neural networks; Original Paper; Pattern Recognition and Graphics; Pixels; Signal,Image and Speech Processing; Vision; Deep learning; Segmentation; Superpixels
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-021-02066-2

Superpixels are to aggregate some pixels with similar characteristics to form a more representative large element. This new element will be the basic unit of other image processing algorithms. It can not only greatly reduce the dimension, but also eliminate some abnormal pixels. Most of the existing superpixel algorithms are not based on deep learning, or only based on simple convolutional neural network. In this method, convolutional neural network with Squeeze-and-Excitation (SE) module is applied to superpixel segmentation, which solve the problem of loss caused by different channel of feature map in convolution pool. The convolution neural network with SE module can segment the image more accurately. In addition, SE nets can be easily integrated into downstream deep networks resulting in performance improvements. The extensive experimental results show that improved disparity estimation accuracy can be obtained on public datasets.