Figure 10

Flow-chart of integrating 4D LF Fourier slice³¹, and combining defocus and correspondence of MRF-depth estimation²⁶ algorithm to calculate 3D morphology image from 2D LF-image. LF-image snapshot on image sensor is 2048 × 2048 pixels; around 19 × 19 directions is used in this analysis; Final depth image is 95 × 95 pixels. (FFT_LF: fast Fourier transfer of LF image array, Refocus_Image: digitally refocused image array, D_α: defocused image array, C_α: correspondence image array, α_D^*: αvalue that maximizes defocus measure, α_c^*: αvalue that minimizes correspondence measure, α_D^*2 & α_C^*2: next local optimal value or next largest peak of defocus and correspondence, D_conf & C_conf: confidence value which is proportional to the ratio of response estimate α^* and α^*2, Z₀: initial array of defocus and correspondence array, W₀: initial array of defocus and correspondence confidence array, Z^*: depth estimation array, λ_source: weighting between defocus and correspondence, λ_flat: Laplacian constraint for flatness of depth estimation, λ_smooth: control of second derivative kernel, δ: difference between Z^* and constraints, є²_softness: softness of next iteration, T: annealing term, and E: error weight matrix)²⁶.