Contribution: 1) Systematic interpretation to existing face sketch synthesis methods. 2) Bayesian face sketch synthesis: apply the spatial neighboring constraint to both the neighbor selection model and the wieght computation model.
Problem:
s代表target patch, t代表test patch,$X_i = {x_k^i}_{k=1}^K和Y_i = {x_k^i}_{k=1}^K$分别为 test patch 的K个最近邻photo patches和sketch patches. target patch 由下面公式计算得到：
$$s_i = Y_i cdot w_i = sum_{k=1}^K w_{ik}y_k.$$
给定test patch生成target patch,等价于最大后验概率$p(s|t) = p(s_1,...,s_N|t_1,...，t_N) = p(W,Y|t) = p(W|Y,t)p(Y|t)$
将上式分为两个部分: P(W|Y,t)和P(Y|t)分别称为weight computation model和neighbor selection model.
Present work:
Neighbor Selection Model: 1) 忽略空间相邻batch的限制，单独考虑每个 text patch 2) 考虑空间相邻的batch限制
Weigth Computation Model: 1) 忽略空间相邻batch的限制，单独考虑每个 text patch 2) 考虑空间相邻的batch限制
MRF is mainly for neighbor selection and MWF is mainly for weight computation.
Bayesian face sketch synthesis:
