Digital Signal Processing Reference
In-Depth Information
and the scale line for position liquid level. Because the rich information of the hori-
zontal edge of the scale and the scale line. In this step, we convolve the bottle image
with horizontal Sobel operators. Edge detection results are shown in Fig.4.
Fig. 4.
Detection results by horizontal Sobel operator
Fig. 5.
Binary image
2.2
An Auto-Adaptive Threshold Binary Image
All milk bottle images are binarized by an auto-adaptive threshold. In Fig.4, we can see
that Gray level higher part gathered at the scale regional component and the overall
share of energy is relatively small. Threshold point in the background and the scale line
is divided by three-quarters of the total energy. Given that H(i) is the gray histogram
value in point i. The following equation is used to obtain optimal threshold value.
T
3
(1)
Hi
[]
≥
4
i
=
0
Fig.5 shows the binary image by above equation. The white part of the image is mainly
distributed in the region of the scale. Therefore, the scale region can be segmented by
this feature.
3
Automatic Segmentation of Scale Regional
From Fig.5, the higher part of the gray value distribution in the vertical direction of the
scale regional. Therefore, we propose projection statistical algorithms for split the scale
regional and non-scale regional. From fig.7 (a), we can see that the larger part of
the statistical value corresponding scale regional. Non-scale area is split by removing
the scale area.
In fact, these larger values corresponding to the larger local maxima. Shen algorithm
[10] is introduced to find the maximum point in the statistical value. The Shen algo-
rithm not only has the function of the low-pass filter, and can calculate the first deriv-
ative and second derivative. Given that I(x) is the original statistical signal and, Flow
diagram are shown in Fig.6.
