Digital Signal Processing Reference
In-Depth Information
the
x
-axis and the True Positive Rate (TPR) on the
y
-axis. The FPR measures the
fraction of negative examples that are detected as positive. The TPR measures the
fraction of positive examples that are correctly labelled. Thus, FPR(
c
) is defined as
FP
/ (
FP
+
TN
) and TPR(
c
) is defined as
TP
/ (
TP
+
FN
).
TP
,
FP
,
TN
and
FN
are
known as True Positive, False Positive, True Negative and False Negative where TP
is examples correctly recognized as positive, FP is negative examples incorrectly
recognized as positive.
TN
corresponds to negative examples correctly identified and
FN
refers to positive examples incorrectly recognized as negative. They are referred
as the four categories of a confusion matrix [24].
Fig. 7.
Receiver Operator Characteristic (ROC) curve analysis for all experimented concepts
with Attention Threshold setting,
t
= 0
In this paper, we intend to evaluate the performance of the proposed model with
respect to the Attention Threshold effect towards the changes of the SFL threshold,
when using different type of feature descriptors. Every concept image dataset as well
as the negative image dataset will be supplied to be recognized by the model, with
different parameters settings. Each case of experiment will produce its unique confu-
sion matrix and represented in ROC curve shown in Figure 7. The area under the
curve in ROC space reveals the performance of the algorithm.
The model achieves almost 1.0 TPR value when recognizing certain concepts, for
example 'Building', 'Human', 'Vehicle' and 'Tiger' by using just CEDD descriptors.
This shows the colour and edge information was being captured and contributing as
