Melanoma and Nevus Skin Lesion Classification Using Handcraft and Deep Learning Feature Fusion via Mutual Information Measures

. 2020 Apr 23;22(4):484. doi: 10.3390/e22040484

Algorithm 1 Algorithm summary

Require: PSL Image

(a)Preprocessing

1:
Input: I
Apply Gaussian Filter, Equation (2)
Apply RGB to CIEL*a*b* Color Transform
Separate CIEL*a*b* Image $I_{L a b}$ in Image Channels $I_{L}, I_{a}, I_{b}$
Calculate Mean value of $I_{L}, I_{a}, I_{b}$ , Equations (3)–(5)
2:
for all $(i, y) \in I_{L}$ do
3:
if $I_{L} (x, y) \geq \bar{I_{L}}$ then
4:
Assign 1 to $I_{T h_{L}} (x, y)$
5:
else
6:
Assign 0 to $I_{T h_{L}} (x, y)$
7:
end if
8:
end for
9:
for all $(i, j) \in I_{a}$ do
10:
if $I_{a} (x, y) \geq \bar{I_{a}}$ then
11:
Assign 1 to $I_{T h_{a}} (x, y)$
12:
else
13:
Assign 0 to $I_{T h_{a}} (x, y)$
14:
end if
15:
end for
16:
for all $(i, j) \in I_{b}$ do
17:
if $I_{b} (x, y) \geq \bar{I_{b}}$ then
18:
Assign 1 to $I_{T h_{b}} (x, y)$
19:
else
20:
Assign 0 to $I_{T h_{b}} (x, y)$
21:
end if
22:
end for
Compute $I_{b i n} (x, y)$ Applying Equation (9) to $I_{T h_{L}}, I_{T h_{a}}, I_{T h_{b}}$
Apply Median Filter, size $= 5 \times 5$
Compute Bounding Box Algorithm to estimate coordinates of Region Of Interest
Crop $I_{r o i} (x, y)$ from estimated coordinates on $I (x, y)$ & $I_{L a b} (x, y)$
23:
Output: Region of Interest (ROI) Image $I_{r o i} (x, y)$

(b)Handcraft Features

24:
Input: $I_{R O I}$
Compute Area, Perimeter, Circularity, Diameter and Eccentricity from Equations (13)–(17)
Compute Asymetry from Equations (10)–(12)
Compute Color Features from Equations (18)–(21) from $I_{r o i} (x, y)$
Compute Texture Features from Equations (22)–(43) from $I_{r o i} (x, y)$
25:
Concatenate the extracted features H
26:
Output: H Handcraft features

(c)Deep Learning features