Focal-UNet: Complex Image Semantic Segmentation Based on Focal Self-attention and UNet
Haosong Gou, Lei Xiang, Xiaonian Chen, Xin Tan, Lei Lv · 2024
Semantic segmentation is an important task in the field of computer vision and is widely used in fields such as medical image analysis and autonomous driving. However, when the recognized subject and the background are very similar and easily confused, it is more difficult to segment the target. This paper proposes an image segmentation algorithm based on Focal self-attention and UNet – Focal-UNet, which uses focal self-attention to capture local and global features and retain the most valuable information for edge recognition. At the same time, data enhancement methods are used to enhance the sensitivity of the model to surrounding local features during edge recognition. Experiments have proven that compared with traditional image segmentation algorithms, our algorithm achieves better segmentation results in scenes with complex backgrounds and high foreground similarities, and improves efficiency without losing more accuracy.