Y
=
P
(
A
(
H
(
X
)
)
)
(
∗
∗
)
Y=P(A(H(X))) \qquad (**)
Y=P(A(H(X)))(∗∗)
其中P为Poisson噪声,X为分子结构图,H为高斯核卷积算子,A为精细网格到粗粒度网格的平均算子。
上述超分问题转化为如下的优化问题
X
^
∈
arg
min
X
1
2
∥
A
H
(
X
)
−
Y
∥
2
+
λ
∥
X
∥
0
+
δ
≥
0
(
X
)
\hat{X}\in\arg\min_{X} \frac{1}{2}\|AH(X)-Y\|^2+\lambda \|X\|_0+\delta _{\ge0}(X)
X^∈argXmin21∥AH(X)−Y∥2+λ∥X∥0+δ≥0(X)
该问题求解为NP-hard
将L0范数替换为Continuous exact
ℓ
0
\ell_0
ℓ0,得到如下模型
X
^
∈
arg
min
X
1
2
∥
A
H
(
X
)
−
Y
∥
2
+
Φ
C
E
L
0
(
X
)
+
δ
≥
0
(
X
)
\hat{X}\in\arg\min_{X} \frac{1}{2}\|AH(X)-Y\|^2+\Phi_{CEL0}(X)+\delta _{\ge0}(X)
X^∈argXmin21∥AH(X)−Y∥2+ΦCEL0(X)+δ≥0(X)
其中
Φ
C
E
L
0
(
X
)
=
∑
i
=
1
(
N
L
)
2
ϕ
(
∣
A
H
(
E
i
)
∣
,
λ
;
∣
X
i
∣
)
\Phi_{CEL0}(X)=\sum_{i=1}^{(NL)^2}\phi(|AH(E^i)|,\lambda;|X_{i}|)
ΦCEL0(X)=i=1∑(NL)2ϕ(∣AH(Ei)∣,λ;∣Xi∣)
这里的各项可表示为
ϕ
(
a
,
λ
;
x
)
=
λ
−
a
2
2
(
x
−
2
λ
a
)
2
1
x
≤
2
λ
a
\phi(a,\lambda;x)=\lambda-\frac{a^2}{2}(x-\frac{\sqrt{2\lambda}}{a})^2 \mathbb{1}_{x\le\frac{\sqrt{2\lambda}}{a}}
ϕ(a,λ;x)=λ−2a2(x−a2λ)21x≤a2λ
Remark:
结合iterative reweighted L1算法2求解CEL0的优化问题。
nature近期提出一种单分子定位的深度学习方法:DECODE
该方法可以快速处理密集emitter的成像数据。
具体见文献3
参考文献:
S. Gazagnes, E. Soubies, and L. Blanc-Féraud, “High density molecule localization for super-resolution microscopy using CEL0 based sparse approximation,” in ISBI, Apr. 2017, pp. 28–31. ↩︎
Peter Ochs et al., “On iteratively reweighted algorithms for nonsmooth nonconvex optimization in computer vision,” SIAM Journal on Imaging Sciences, vol. 8, no. 1, 2015. ↩︎
A. Speiser et al., “Deep learning enables fast and dense single-molecule localization with high accuracy,” Nat Methods, vol. 18, no. 9, pp. 1082–1090, Sep. 2021. ↩︎