Defect Detection in Fabrics Using Local Binary Patterns - Advances in Image and Graphics Technologies - page 281

To compare the detection accuracy of different parameters of the LBP, false detec-
tion rate is proposed as a measure scale to quantify the LBP operator. It is defined as
follows:

N + N

E = (

c d

) × 100 %

(10)

N

total

Where, E represents the false detection rate. The smaller the value is, the better the
detection algorithm does. N c is the number of the windows which are detected as defects
windows while they are defects-free windows. N d means the number of the windows
which are detected as defects-free windows while they are defects windows. N total shows
the total number of the windows which are got by dividing the defect image.
According to Eq.10, the false detection rate of the 10 images processed by LBP 8, 1 ,
LBP 16, 2 and LBP 24, 3 can separately be calculated, and the results are carried on an
analysis, which are shown in Table 1.

Table 1. The false detection rate of the 10 images processed by LBP 8, 1 , LBP 16, 2 and LBP 24, 3
separately (%)

LBP 8 1 LBP 16 2 LBP 24 3 Average

Defect image

C1R1E2CH (A) 4.21 3.89 3.17 3.76

CIR1EAD (B) 4.92 4.38 4.07 4.46

C1R1E2N1 (C) 5.28 4.52 4.06 4.62

C1R2EBOA (D) 4.83 4.27 4.42 4.48

C4R1CLOCH1 7.04 3.85 4.36 5.08

C2R1ECP (F) 2.89 4.74 5.97 4.53

C2R2ECXO (G) 3.32 4.73 6.54 4.86

C2R2E1N4 (H) 6.69 6.13 5.08 5.97

C3R2EDUZ (I) 5.74 4.86 3.51 4.70

t1 (J) 4.54 3.98 3.27 3.93

Average 4.95 4.54 4.45

From Fig.4 and Table 1 it can be visually seen that the defect detection effect of
LBP 24,3 operator is the most accurate. The average of false detection rate is 4.45%.
When processing the defect images with LBP 16,2 operator, there are small parts of the
leakage inspection windows existing by mistake. The average of false detection rate is
4.54%. While the average false detection rate is 4.95% when applying LBP 8,1 operator.
The error detection section will be more than applying LBP 16, 2 operator. However, with