XTC Texture Classifier
A machine learning model for identifying similar sub-images using VGG16 and linear regression.
The XTC Texture Classifier is part of a broader research effort in image super-resolution, aiming to identify high-quality sub-images similar to a low-quality input. The system uses the VGG16 model, a pre-trained convolutional neural network, to extract high-level features from sub-images. These features are passed through a lightweight linear regression model, which computes similarity scores to find matches between low-resolution inputs and high-resolution target sub-images.
To align with its intended use case, the training process involves scaling down input images to mimic low-quality conditions. The model is trained on the DIV2K dataset, leveraging its high-resolution images to capture intricate texture patterns. By learning to map low-quality inputs to their high-quality counterparts, the classifier achieves a 95% accuracy rate.