MUSE: A Multi-view Synthesis Enhancer

In this paper, we introduce the MUSE (MUlti-view Synthesis Enhancer) method, which is an evolution of our previously proposed HDSB method and is based on a hybrid algorithmic-learning-based scheme. MUSE generates novel views of a scene using an autoencoder specifically optimized for refining pre-synthesized views that have been derived from actual observations. Since the subjective test is the ultimate test of the visual rendering quality, we evaluate our proposed method by two subjective tests. Experimental results show that the MUSE brings a global gain compared to two tested state-of-the-art methods.