Interactive Segmentation of Art Images based on Point Cloud Pre-Training

This study presents an innovative interactive segmentation algorithm for fine art images using point cloud pre-training principles. With increasing demands for precision and variety in image processing, conventional methods are proving inadequate. Image segmentation, a central aspect of visual data manipulation, involves dividing an image into distinct regions based on static or dynamic features. This study reviews supervised and unsupervised segmentation strategies, emphasizing the importance of fuzzy clustering algorithms in practical applications. Despite advances in medical image segmentation, challenges remain due to the complex nature of image processing. A universally applicable segmentation method and standardized evaluation criteria are elusive. To address these challenges, the proposed algorithm combines Local Binary Pattern (LBP) feature extraction with a point cloud pre-training model, utilizing the U-net architecture for accurate segmentation. Experiments conducted on an Ubuntu 14.04 64-bit OS using C++ and Python demonstrate the effectiveness of the algorithm. The visualized results show its potential for practical applications and lay the foundation for future advances in image segmentation.