Image super-resolution via feature-augmented random forest

Recent random-forest (RF)-based image super-resolution approaches inherit some properties from dictionary-learning-based algorithms, but the effectiveness of the features working in RF is overlooked in the literature. In this paper, we present a novel feature-augmented random forest (FARF) method for image super-resolution, where the conventional gradient-based features are proposed to augment the features used in RF, and different feature recipes are formulated on different processing stages in an RF. The advantages of our method are that, firstly, the dictionary-learning-based features are enhanced by adding gradient magnitudes, based on the observation that the non-linear gradient magnitudes are highly discriminative. Secondly, generalized locality-sensitive hashing (LSH) is used to replace principal component analysis (PCA) for feature dimensionality reduction in constructing the trees, but the original high-dimensional features are employed, instead of the compressed LSH features, for the leaf-nodes’ regressors. With the use of the original higher dimensional features, the regressors can achieve better learning performances. Finally, we present a generalized weighted ridge regression (GWRR) model for the leaf-nodes’ regressors. Experiment results on several public benchmark datasets show that our FARF method can achieve an average gain of about 0.3 dB, compared to traditional RF-based methods. Furthermore, a fine-tuned FARF model can compare to, or (in many cases) outperform, some recent state-of-the-art deep-learning-based algorithms.