FADNet: A Feature-Adaptive Framework for Furniture Decoration Image Segmentation

As the furniture decoration industry continues to evolve and innovate to meet consumer demands for personalized decor, precise and efficient segmentation and processing of furniture decoration images are essential. Currently, neural network algorithms are widely used for furniture decoration image segmentation. However, these methods struggle to accurately learn image feature weights, resulting in suboptimal segmentation outcomes. To address this, we propose a feature-adaptive framework for furniture decoration image segmentation. First, texture features of furniture decoration images are extracted using self-supervised learning algorithms based on image storage and pixel distribution types. Second, color space selection is employed to refine image contours using an active boundary loss coefficient. A locally adaptive model is introduced to learn image feature weights effectively. Finally, a Vision Transformer is utilized to construct an image segmentation model, achieving superior segmentation of furniture decoration images. Experimental results on our Furniture Decoration dataset demonstrate that the proposed method effectively segments target areas in images, with a Dice coefficient reaching up to 0.913. The proposed method also achieves a recall of up to 0.9, indicating its practical utility for complex indoor scenarios.