Semantic Segmentation of Facial Wrinkles Based on Axial Convolution and Wavelet Downsampling

With the widespread application of deep learning technology in the field of medical image analysis, efficient and accurate facial wrinkle detection has become a challenging task. This study proposes a facial wrinkle semantic segmentation method that combines axial convolution with wavelet downsampling. The proposed method aims to enhance wrinkle detection accuracy and efficiency by merging deep learning with advanced image processing techniques. An axial convolutional network is designed for effective extraction of wrinkle details and directional features. This network processes images independently along the x-axis and y-axis. Wavelet downsampling is utilized to optimize feature retention during the downsampling process. Experiments show that our method has significant advantages in wrinkle detection. Compared with UNet, the IoU is increased by 20.05% and the Dice score is increased by 20.54%. This achievement provides a new technical approach for the automatic detection of facial wrinkles, and also provides an important reference for future research in the field of dermatology diagnosis and beauty.