Automatic Classification of Human Embryo Microscope Images Based on LBP Feature - Advances in Image and Graphics Technologies - page 149

3 Experiment and Analysis

3.1 Feature Selection

We extracted two kinds of features on 185 embryos images from Assisted Reproduc-
tive Center of Navy General Hospital, PLA (the number of samples that are success-
fully pregnant is 47, the number without success to conceive is 138) has been used to
make classification experiments. The first kind of feature is the five order central
moments of embryo images, namely using each first five central moments in horizon-
tal and vertical direction to classify embryos. The central moments contain texture -
related information of the change intensity of images, and they can be used to classify
the embryo images[3]. The corresponding central moment vector of each sample is
12-dimensional. The second kind of feature is the LBP feature. The LBP operator can
be used to measure and extract local texture information of images, and it is illumina-
tion invariant. In this experiment, the LBP vector of each sample is 256-dimensional.
To verify the classification algorithm, we used six cross-over trials to design the
classifier. Data were randomly divided into six groups. One group was chosen as the
test data each time, and the remaining five groups were employed to train the classifier.

3.2 Classifier Design and Analysis

3 . 2.1 Central Moment
When all five central moments of the original images are used to do classification, the
support vector indexes are 2,5,7,171, respectively. Based on the projection direc-
tion w that is [0.0009,0.0001,0.0007,0.0136,0.0619,-0.0043,0.0009,0.0002,0.0018,
0.0379,0.1684,-0.0036]×10 -3 the b ∗ values corresponding to the four support

vectors are all -2.5384. So the final decision function took the offset b as -2.5384. We
used this classifier to classify the embryo images and the average accuracy rate was
62.16%. Fig. 2 shows the projection result of samples to one-dimensional space

aprojection result bthe expansion of figure (a) near the origin

Fig. 2. The adding result of samples' projection onto the classification plane's normal vector
and offset for central moments