Texture Classification Using ResNet and EfficientNet

The texture is the fundamental unit of an image that aids in its identification. Analysis of texture forms the foundational basis of Computer Vision tasks like image recognition and image segmentation. Images of various domains like satellite and medical have been identifiable because of their texture. This paper aims to propose models for the task of texture classification, evaluate them on a standard data set and then compare the results of the models to the results attained by models proposed previously for the same task. The paper uses the transfer learning approach to achieve this objective. The pre-trained models are used for feature extraction, removing their top layer and freezing the rest. The two pre-trained models used in this paper are the ResNetV2 and the EfficientNet-B4 paper. The proposed models are trained and tested on the Kylberg data set, a widely used texture data set. The two models attained accuracies of 99.78% and 92.97%, respectively.