Compressively sampled light field reconstruction using orthogonal frequency selection and refinement

This paper considers the compressive sensing framework as a way of overcoming spatio-angular trade-off inherent to light field acquisition devices. We present novel method reconstruct full 4D from sparse set data samples or measurements. The approach relies on assumption that models in Fourier domain can efficiently represent fields. proposed algorithm reconstructs fields by selecting frequencies basis functions best approximate available hyper-blocks. performance reconstruction is further improved enforcing orthogonality approximation residue at each iteration, i.e. for selected function. Since sparsity better preserved continuous domain, we propose refine searching neighboring non-integer frequency values. Experiments show yields improvements more than 1 dB compared state-of-the-art methods. refinement step also significantly enhances visual quality results our 1.8 average.