Loss-Specific Training of Random Forests for Super-Resolution

Super-resolution addresses the problem of image upscaling by reconstructing high-resolution output images from low-resolution input images. One successful approach for this problem is based on random forests. However, this approach has a large memory footprint, since complex models are required to achieve high accuracy. To overcome this drawback, we present a novel method for constructing random forests under a global training objective. In this way, we improve the fitting power and reduce the model size. In particular, we combine and extend recent approaches on loss-specific training of random forests. However, in contrast to previous works, we train random forests with globally optimized structure and globally optimized prediction models. We evaluate our proposed method on benchmarks for single image super-resolution. Our method shows significantly reduced model size while achieving competitive accuracy compared to state-of-the art approaches.