Open Journal of Applied Sciences, 2012, 2, 267-271
doi:10.4236/ojapps.2012.24039 Published Online December 2012 (http://www.SciRP.org/journal/ojapps)
The Research of Contrast Enhancement Algorithm
in Laser Projection Display System
Bailin Na, Yingying Wu
School of Information Science and Technology, East China Normal University, Shanghai, China
Received September 30, 2012; revised October 29, 2012; accepted November 10, 2012
ABSTRACT
High-contrast is one of the main advantages in laser projection display, and the method of DCC (Dynamic Contrast
Control) is the main way to increase the contrast. Generally, image pre-processing is necessary for eliminating noise
and decreasing the over-highlight. In this paper, we proposed and actualized a method by following 3 steps: Firstly, the
original image was analyzed statistica lly to get th e scope of gray-scale distribution and average gray-scale; and then the
image was divided into a number of sub-images. The sub-images whose pixels are higher than a certain threshold in
both number and range, are applied image segmentation by certain growth rules. The sub-images satisfied with the
growth rules are marked 1, and the rests are marked 0. Secondly, the sub-images are uniting. A sub-image has 3 rela-
tions between 8 sub-images around it: 1 and 1, 1 and 0, 0 an d 0. The sub-images marked 1 are uniting to gether, and the
sub-images marked 0 are uniting together. Without affecting the visual vision, all over-highlight pixels were reduced in
a certain proportion. Lastly, based on the app lication of DCC, the whole image signals were enlarged and the brightness
of light sources were reduced, so as to achieve the desired effect in contrast enhancement.
Keywords: Contrast; Region-Grow Segmentation; Edge-Tracking
1. Introduction
Liquid crystal on silicon (LCOS) is a “micro-projection”
or “micro-display” technology typically applied in pro-
jection televisions. It is a reflective technology similar to
DLP projectors; however, it uses liquid crystals instead
of individual mirrors. By way of comparison, LCD pro-
jectors use transmissive LCD chips, allowing light to
pass through the liquid crystal. In LCOS, liquid crystals
are applied directly to the surface of a silicon chip coated
with an aluminized layer, with some type of passivation
layer, which is highly reflective.
A key metric of image quality for a projected or di-
rectly-viewed image is the contrast ratio. Contrast ratio
itself consists of two measurements, “on/off contrast”, or
full-screen contrast, and ANSI contrast which uses a field
of 16 black and white rectangles. ANSI contrast can be
used to describe the influence of light scattering on the
display, and ANSI contrast cannot exceed the value of
full-screen contrast. For computer graphics displays,
where images have large areas of white or other bright
colors displayed, the ANSI contrast value is a very useful
metric. An ANSI contrast value of 300:1 is usually con-
sidered sufficient due to the dynamic range limitations of
the human eye.
2. Projection Display Principle
In the projection display system, the mechanism of pro-
jection display can be equivalent to the formula:
*
M
SI

(1)
where M stands for the final image projected on the
screen, S stands for the signal intensity of image, and I
stands for the luminance of light source. In conventional
systems, the DCC is used to enhance the contrast to im-
prove the image visual effects. The DCC increase the
signal intensity, and reduce the luminance of light source
simultaneously. So the interferences of parasitic light and
diffractive light are also be reduced too. Just as the for-
mula:
 
1
*MS I


1






. (2)
Histogram modification based algorithm is the most
popular approaches to achieve widely dynamic range.
Histogram Equalization (HE) is one of the most com-
monly used algorithms to perform contrast enhancement
due to its simplicity and effectiveness. In general, the HE
distributes pixel values uniformly and results in enhanced
images with linear cumulative histogram. But there are
many disadvantages that HE enhances the entire image
pixels and the visual results are hard to control, and most

B. L. NA, Y. Y. WU 268
of all is that the noises are enhanced too.
So in this paper, we proposed and actualized a new
system based on the principle of image segmentation.
3. Image Segmentation
Image segmentation is one of the most important tasks in
image processing, and has a wide range of applications in
computer vision, such as pattern recognition, image
compression and so on. The image segmentation ap-
proaches can be divided into four categories: threshold-
ing, clustering, edge detection and region extraction. In
this paper, a region extraction based method for image
segmentation will be considered, the special regions
which has a strongly contrast to its surroundings will be
gotten by using image segmentation [1].
We detect the edges firstly by using the templates
which have (2P + 1)*(2P + 1) pixels in size. If the center
pixel of the template is located in non-edge regions, then
the whole gray-scale changes within the template should
be gentler. In other words, the gray value of center pixel
should be close to the pixels which are located around it.
If the center pixel in the template is located in edge re-
gion, then the gray-scale changes in the template region
should be relatively intense.
To achieve the above purposes, we definite a variable

to reflect the standard deviation of the changes in
local regions of gray-scale image, for a given image g
(i,j). With the local gray-scale change from moderate to
severe, the

would increase in correspond [2]. We use
the template P to mean-filter the whole image, to calcu-
late the mean of local pixels (mg) ,and the standard devia-
tion of local pixel s




by the formulas:

 
2
1
,,
21
ip jp
gkiplj p
mij gkl
p

 


(3)
   
2
2
1
,,
21
ip jp
g
kipl jp
ijgklm kl
p


 



 ,
(4)
Figure 1(a) is the original image, and th e Figure 1(b)
is processed image. In Figure 1(b), we can see the obvi-
ous edge of local region and noise. So the next step, we
use Gauss Filter and Median Filter to eliminate the dis-
turbances of noises. After that we need to choose a
threshold which is very important in the step to keep the
high contrast edges and ignore the low contrast edges. In
this paper, a large number of images are analyzed statis-
tically to get the scope of the thresho ld in that the choice
of the threshold will be good to eliminate interference.
The image below has been processed with the threshold.
Figure 2(b) is the processed image. We can see that
most of noises are eliminated and the high-contrast edges
are strengthened. The next step we need to estimate
whether that there are closed regions in the image or not.
If there are closed regions, then we continue to work. If
not, we do not process the image anymore. We estimate
it in this way:
The method to estimate closed regions is edge-track-
ing. We use counterclockwise edge-tracking in accor-
dance with the order of the arrows in Figure 3(a). At the
same time; we mark the pixels every time to prevent in-
OriginalM EdgeM
Figure 1. The original image and processed image.

B. L. NA, Y. Y. WU 269
Thre sh ol dM EdgeM
Figure 2. The edge image.
i,j
j
1
2 3 4
5
8 7 6
i,j
Figure 3. Growth rule.
finite loops. We track the edge according to the direction
of the arrow. Part of the code is as follows:
if (M (i,j + 1) = threshold) &(State! = 5)
mask (i,j + 1) = 1;
i = i + 1;
j = j + 1;
State = 1.
4. Region-Growing
Region-growing approaches exploit the important fact
that the pixels which are close togeth er have similar gray
values. In region-growing process, there are two factors
must to be considered, the first one is how to choose the
seed(s) in practice, and the second one is how to choose
the similarity criteria. The method of selecting the seeds
is a key step to segmentation, because the segment result
is sensitive to the selection of the initial growing points.
For example, the result of region growing will go awry if
the initial seed falls on a noise point. For a good seg-
mentation, it is required that the regions have relative
uniform gray value and the seed pixels have a gray value
which is typical of the region. In this paper, we will make
some improvement on the method of selecting seeds and
region growing rules [3 ].
The method we proposed:
1) We analyze the entire image pixels statistically, and
then calculate the average gray of the image. In this step,
we need to get the threshold which can be calculated
based on adaptive algorithm, and the threshold would be
applied in the next steps.
2) We divide the original image into a number of
sub-images which are in the same size. (The number of
sub-images would affect the image segmentation). The
gray-scale of the sub-images is analyzed statistically. The
pixels which are greater than the threshold are called
high-light pixels, and the number of high-light pixels of
every sub-image is counted. If the high-light pixels num-
ber of a sub-image is larger than the stipulated limit, the
sub-image should be marked 1, and process the region-
growing in the next step. If not, the sub-image should be
marked 0, do nothing [4].
3) The sub-images marked 1 are processed with re-
gion-growi ng algorithm.
The seeds of region-growing: we analyze the sub-im-
age statically to get the seeds.
The rules of region-growing: the seeds grow anti-
clockwise among the 8 pixels aro u nd it.
4) The last step was region uniting. A sub-image has 3
relations between 8 sub-images around it: 1 and 1, 1 and
0, 0 and 0. So the relationship between the sub-images
could be concluded in three kinds. In the first two kinds
of situation, the relationship between the pixels at the

B. L. NA, Y. Y. WU 270
sub-image edge is used to determine whether to uniting
or not. In this step, if the sub-images which are marked 0
have been uniting with others, they would be marked 1
after uniting. Finally, in the k ind of 00, if the p ix els at the
edge between two sub-images meet the growth rules, the
two sub-images are uniting, and are marked 1. The
sub-images which are still marked 0 after segmentation,
are considered as noise and would be eliminated
Figure 4(a) is the effect image after image segmenta-
tion.
5. Result and Conclusions
To compare and evaluate the processing result, we do
some experiments on general photo image.
Figure 5(a) is the original image. Figure 5(b) is proc-
essed image. Without affecting the visual vision, the re-
gions of all over-highlight pixels were reduced in a cer-
tain proportion.
OriginalM
RegionM
Figure 4. The highlight areas.
OriginalM ProcessedM
Figure 5. The original image and the final image.

B. L. NA, Y. Y. WU 271
OriginalM
0 50 100 150 200 250
0 50 100 150 200 250
4000
3500
3000
2500
2000
1500
1000
500
0
4000
3500
3000
2500
2000
1500
1000
500
0
ProcessedM
Figure 6. The HEs of the original image and the final image.
Figures 6(a) and (b) are the histog rams of the original
image and the processed image. We can see that the main
parts of the two wave shapes have not changed, that
means the backgrounds of the image stay the same. And
the waveform of the highlight regions have not changed
but moved, that means the highlight regions have been
reduced. For example: If the original image could be
enlarged 1.2 times, then the processed could be enlarged
1.3 times:

1
1.3** 1.3
M
S



I


(5)
Lastly with the application of DCC, the whole image
signal was enlarged and brightness of light source was
reduced, so as to achieve the desired effect in contrast
enhancement.
6. Acknowledgements
Our research is supported by the National High Tech
Research and Development Program of China (2007AA-
030112 and 2009AA032708) .
REFERENCES
[1] S.-Y. Wan and E. H. William, “Symmetric Region Grow-
ing,” IEEE Transactions on Image Processing, Vol. 12,
No. 9, 2003, pp. 1007-1015.
doi:10.1109/TIP.2003.815258
[2] Y. S. Frank and S. X. Cheng, “Automatic Seeded Region
Growing for Color Image Segmentation,” Image and Vi-
sion Computing, Vol. 23, No. 10, 2005, pp. 877-886.
doi:10.1016/j.imavis.2005.05.015
[3] L. T. Law and Y. M. Cheung, “Color Image Segmenta-
tion Using Rival Penalized Controlled Competitive Learn-
ing,” Proceedings of the International Joint Conference
on Neural Networks, IJCNN’03, Portland, 20-24 July
2003, pp. 108-112. doi:10.1109/IJCNN.2003.1223306
[4] R. Pohle and K. D. Toennies, “Segmentation of Medical
Images Using Adaptive Region Growing,” Proceedings
of SPIE—The International Society for Optical Engineer-
ing, Vol. 4322, No. 3, 2001, pp. 1337-1346.