Hybrid CNN–Fuzzy Framework for Dress Pattern Recognition

The burgeoning fashion and textile industry coupled with the rise of e-commerce platforms and digital commerce solutions has raised the demand for automated and intelligent solutions for analyzing garments. Among the various features in an image, dress pattern recognition is a significant part of product binarization or grouping, textile inspection, fashion analytics solution development, and the visual search functionality. However, the automatic recognition of dress patterns from images is a challenging task because of the lighting conditions in the images, the corresponding textures in the garments, the presence of wrinkles in the clothes, the background noise in the images, and resemblance in the different categories of patterns. This research work brings forth an automated dress pattern recognition system using deep learning and fuzzy logic for accurate and interpretable results. For this purpose, the proposed system utilizes VGG16 transfer learning with a focus on a deep learning-based feature extraction architecture that utilizes the pre-trained ImageNet weights to effectively capture intricate texture and pattern-level traits from fabric image inputs. To boost the quality of inputs and improve robustness, a robust image processing and pre-processing scheme is also developed that comprises denoising, brightness normalization, contrast enhancement, normalization, and resizing. Finally, the extracted features.