Traditional Indian Textiles Classification using Deep feature fusion with Curvelet transforms

With the increasing demand for online shopping in the competitive fashion market, new designs come as early birds, and fabric texture plays a crucial role in selecting the correct fabric design. The Indian traditional textile patterns are varied and vibrant, and it exhibits the culture of the area of its origin. They are widely in demand in the international market. Unfortunately, with the incursion of mechanization and the low yield of handmade textiles, artisans for such art forms are dwindling. Generating new designs in sync with the market stream is time-consuming, and the art must be learned meticulously. Hence, every single attempt at developing technology to conserve these art forms is the moment’s need. This paper proposes a Traditional Indian Textiles Classification based on a fusion of deep features and Curvelet transforms. The traditional motif images are complex. Rather than straight lines, it contains curves; hence, the curvelet transform is designed to handle it. Compared with other transforms, Curvelet transforms allow a more systematic representation of other singularities and edges along lines. This work was tested with the Indian art forms datasets. It utilized the pre-trained CNN architecture (InceptionresNetV2, VGG16) as a feature extractor and concatenated these features with curvelet features. In this experiment, the XGB classifier provided the best results (precision 98.24%, recall 97.15%, F1Score 97.15%, accuracy 97.15%, and specificity 99.52%) with 4scale Curvelet and InceptionResNetV2 feature sets. These initial results are promising and motivate further work on larger and more complex datasets.