2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia

Contrast Enhancement for Ziehl-Neelsen Tissue

Slide Images using Linear Stretching and
Histogram Equalization Technique
M.K. Osman #1, M.Y. Mashor #2, Z. Saad *3, H. Jaafar +4
# 1,2
Electronic & Biomedical Intelligent Systems (EBItS) Research Group,
School of Mechatronic Engineering, Universiti Malaysia Perlis,
02600 Jejawi, Arau, Perlis, MALAYSIA.
1
[email protected]
2
[email protected]
*3
Faculty of Electrical Engineering, Universiti Teknologi MARA (UiTM) MALAYSIA,
Kampus Pulau Pinang, Jalan Permatang Pauh
13500 Permatang Pauh, Pulau Pinang, MALAYSIA.
3
[email protected]
+4
Department of Patology, School of Medical Science,
Universiti Sains Malaysia,
16150 Kubang Kerian, Kelantan, MALAYSIA.
4
[email protected]

Abstract—Ziehl-Neelsen staining method is a common Currently, the process in preparing the slide is carried
technique used to diagnose tuberculosis infection. Clinical out manually by the technologists and consists of several
samples such as sputum or tissue are stained using Ziehl- procedures which need to be done very carefully.
Neelsen procedure and analysed manually by microbiologist Therefore, there are cases whereby the slide is not of the
for detecting TB bacilli. However, the screening process is same thickness or the slide has been over stained. This
labour intensive and time consuming. Image enhancement will cause the medical image is often with lower contrast,
tools can be used to improve the manual screening process. darker colour and faint characteristics when it is being
This paper analyses the performance of linear stretching analysed under microscope.
and histogram equalization techniques for Ziehl-Neelsen In medical situation, clarity of images is the main
tissue slide images. Both enhancement techniques which are factor that influenced the accuracy of interpretation.
originally implemented for gray-scale image have been Therefore, image enhancement is a necessary task to
adapted for colour images. Although the adapted image produce fine images by eliminating poor contrast and
processing technique is simple, the results indicate that these unwanted information.
methods may have some potential to be used for improving
the quality of Ziehl-Neelsen slide images. Overall analyses There are a few studies on computer assisted TB bacilli
show that the linear stretching using local minimum detection in sputum using image processing algorithms.
and maximum value has successfully improved the image Veropoulous et al. [2, 3] proposed an automated
identification of TB bacilli system using image
contrast and enhance the image quality.
processing and neural networks. The study implemented
Canny edge detection, region labelling and removal, edge
Keywords — Contrast enhancement, TB bacillus detection, pixel linking, and boundary tracing in image processing
tissue section, linear stretching, histogram equalization. step. Similarly, Forero et al. [4] also proposed to use
Canny edge detection, followed by non-maxima
I. INTRODUCTION suppression and a hysteresis threshold operations.
The Ziehl-Neelsen (ZN) stain, also known as the acid- Unfortunately, very little research has been carried out
fast bacilli (AFB) stain technique is a common technique to analyse the TB bacilli in tissue section [5, 6].
and widely used to identify the presence of tuberculosis Identification of TB bacilli in tissue is more challenging
(TB) bacilli in clinical samples. This is because the compared to sputum. TB bacilli in tissue are not clearly
method is simple, rapid, low cost and effective [1]. visible due to the complexity of background in tissue
For pulmonary TB, sputum sample taken from compared to sputum. The TB bacilli appear in faint,
suspected TB patient, are stained using ZN procedure. concealed and occluded by cells.
The procedure will make the TB bacillus appear red For this reason, the current study investigates the
while other cells and organisms in the sputum smear effectiveness of linear stretching and histogram
sample will remain blue background. For extra- equalization in enhancing Zeihl-Neelsen tissue slide
pulmonary TB, histology of tissue samples obtained by images. Both methods are the simplest and most effective
biopsy or surgery can confirm the diagnosis of tuberculosis techniques for gray-level image enhancement. However
in the majority of cases [1]. with some adaptations, this paper extended these methods
for colour image enhancement. The advantages of

978-1-4244-4683-4/09/$25.00 ©2009 IEEE 431

 2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia

enhancing the slide image is to enable the TB bacilli in the Two methods have been proposed to determine the
Ziehl-Neelsen tissue slide image to be easily identified by minRGB and maxRGB values. The first method is by using
the technologists and to ease the process of image its local minimum and maximum value of each RGB
segmentation in the later part of a diagnosis system.
component separately, and the second method is based on
II. METHODOLOGY average of minimum and maximum value of RGB, as
formulated in Equation 3.
The research methodology involves the following key
tasks: 1
minRGB = minR +minG minB [3]
A. Image Acquisition 3
The Ziehl-Neelsen slide images for tissue section were 1
analysed using Leica microscope at 40x magnification. maxRGB = maxR +maxG +maxB [4]
Then, Infinity 2 camera was used to capture images and 3
saved into bitmap (*.bmp) format with the resolution of where minR, minG and minB are the minimum value of
800x600. each colour component, while maxR, maxG and maxB are
the maximum value of each colour component,
B. Colour Image Enhancement respectively.
Some major difficulties faced in image processing are
contrast and brightness problems. To overcome these,
image enhancement is employed to increase the contrast 2. Histogram equalization
of an image. In general, an image can be enhanced by Histograms are the basis for numerous spatial domain
spreading the range of colour values to make use of all processing techniques. The gray-level histogram of an
possible values. This method is called contrast stretching. image is the probability of occurrence of each grey level
It changes the distribution and range of the digital in the image. Histogram of an image can be manipulated
numbers assigned to each pixel in an image. This is for image enhancement task [8]. One of the common
normally done to accent image details that may be techniques is called histogram equalization. Histogram
difficult for the human viewer to observe. Two types of equalization is a histogram specification process which
contrast stretching methods, linear stretching and consists of generating an output image with a uniform
histogram equalization have been utilised for colour histogram distribution. The purpose of equalizing a
image enhancement. histogram is to stretch and redistribute the original
histogram using the entire range of levels of the image, so
that an enhancement of image contrast is achieved [9].
1. Linear stretching
Let , 0,1,2, … 1 represent the histogram
In general, the linear stretching method for gray-level associated with the intensity levels of a given image, and
image is based on linear mapping function used to the values in a normalized histogram are approximations
enhance the brightness and contrast level of the images. to the probability of occurrence of each intensity level in
Consider the mapping function as given in [7]: the image. For discrete transformation function, the
histogram equalization of an image is given by:
Pin -min k
Pnew = fmax - fmin +fmin [1]
max-min
sk =T rk = pr rj [5]
where
j=1
Pnew : new pixel level k
Pin : input pixel level nj
= k=1,2…L-1
min : lowest pixel value in the image n
j=1
max : highest pixel value in the image
fmin : minimum value of desired range
fmax : maximum value of desired range for k = 0,1,2,…,L-1, where sk is the intensity value in the
processed image corresponding to value rk in the input
For linear stretching process, the value of fmin and fmax are image, n is the total number of pixels, and nj is the total
set to 0 and 255 respectively. number of pixels having gray-level rj.
In colour image, two methods have been proposed to
In colour image, linear stretching of each RGB
histogram-equalize the image. The first method is by
component is carried out separately. Hence Equation 1 performing histogram equalization separately for each
can be further interpreted as follows: RGB component. For the second method, a mean
histogram is formed based on the average histogram of R,
Pin,RGB -minRGB G and B components. Then the mean histogram is used to
Pnew,RGB = ×255 [2]
maxRGB -minRGB perform histogram equalization for each RGB
component.
where
Pnew, RGB : new pixel color value III. RESULTS AND DISCUSSION
Pin, RGB : input pixel color value In order to evaluate the effectiveness of these methods,
minRGB : lowest pixel value in the image 100 images were captured from a set of four positive
maxRGB : highest pixel value in the image control TB slides. The slides are prepared by the
Pathology Department, USM Hospital, Kelantan.

432
 2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia

a.

Figure 1. Slide image with low density of TB bacilli

Figure 1 shows a captured image with low TB bacilli

density. The arrows indicate the TB bacilli in the image.
The TB bacilli are hardly be seen because of lower
contrast and concealed by cells. The images are being test
using four techniques mention previously. Figure 2 shows
the resultant images after applying these methods.
Generally, it can be seen that by applying linear
stretching and histogram equalization have slightly
increased the contrast of TB bacilli images. The TB
bacilli which are magenta in colour also appear clearer. b.
Histogram equalization using separate RGB
components has obviously enhanced the contrast and
suppressed some unwanted staining region, followed by
histogram equalization using mean histogram and linear
stretching using local minimum and maximum value.
However, the TB bacilli contrast seems to increase a little bit
when applying linear stretching using average minimum and
maximum value compared to its original image.
The same results have also been achieved for slide image
with high density of TB bacilli. Figure 3 shows an image
with high density of TB bacilli. There are numerous TB
bacilli scattered over the entire image. Figure 4 shows the
resultant images after applying the proposed methods. It
is shown that histogram equalization using separate RGB
components produced the best enhancement compared to c.
other techniques. On the other hands, there is a slight
increase of TB bacilli contrast when applying linear
stretching using average minimum and maximum value
compared to its original image.
Finally the performances of the proposed methods are
tested using over staining image. Figure 5 shows a
captured TB bacilli image with over staining problem.
The TB bacilli are hardly being seen since the staining
colour is much similar to the TB bacilli. Figure 6 shows
the resultant images after applying the proposed methods
In this cases, it is seems that linear stretching using
local minimum and maximum value produced acceptable
results, followed by linear stretching using average
minimum and maximum value. Over enhancement occurs
when applying both histogram equalization using d.
separate RGB histogram and mean histogram. The over
Figure 2. Results of applying a. linear stretching using local
staining region became darker and reduced the TB bacilli minimum and maximum value b. linear stretching using average
contrast. minimum and maximum value c. histogram equalization using separate
RGB histogram and d. histogram equalization using mean histogram

433
2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia

d.
Figure 4. Results of applying a. linear stretching using local
minimum and maximum value b. linear stretching using average
minimum and maximum value c. histogram equalization using separate
Figure 3. Slide image with high density of TB bacilli RGB histogram and d. histogram equalization using mean histogram

a.

Figure 5. Over-stained TB tissue slide image

IV. CONCLUSION
In this paper the results of applying linear stretching
and histogram equalization for TB slide images in tissue
have been presented. Through the experiments using the
test images having different characteristics, it has been
proved that linear stretching using local minimum and
maximum value produced acceptable results and visually
comfortable image compared to other techniques. Even
though histogram equalization using separate RGB
b. histogram contributed the best contrast enhancement in
several situation compared to other methods, it failed
when applied to over staining slide images. These
enhanced slide images can be further used for TB bacilli
detection by the technologists and also ease the process of
image segmentation in the later part of a diagnosis system.

ACKNOWLEDGMENT
The authors wish to thank the panel of pathologist
from the Department of Pathology, School of Medicine,
University Science Malaysia, who has participated in this
study and permission to carry out work. Furthermore, the
authors are thankful to MOSTI for providing the grant
money.
c.

434
 2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia

REFERENCES on Biomedical Engineering 2008 BIOMED 2008, Kuala Lumpur,

Malaysia 2008, pp. 583-586.
[1] N. A. Khaled and D. A. Enarson, Tuberculosis-A Manual for [6] S.I. Sahidan, Zaleha Salleh, M. Y. Mashor, A.S.W. Wahab, and
Medical Students. Geneva: World Health Organization, 2003. H. Jaafar, "Contrast Image Enhancement in Tissue Section for
[2] K. Veropoulos, C. Campbell, and G. Learmonth, "Image Detecting Tuberculous Bacilli," in Malaysian Technical
processing and neural computing used in the diagnosis of Universities Conference on Engineering and Technology
tuberculosis," in IEE Colloquium on Intelligent Methods in (MUCET 2008) Perlis, Malaysia, 2008.
Healthcare and Medical Applications (Digest No. 1998/514), [7] J. R.W. Weeks, Fundamentals of Electronic Image Processing:
York, UK, 1998, pp. 8/1-8/4. Bellingham: SPIE Press, 1996.
[3] K. Veropoulos, C. Campbell, Learmonth, G., , B. Knight, and J. [8] R.C. Gonzalez and R. E. Woods, Digital Image Processing, 3
Simpson, "The Automated Identification of Tubercle Bacilli ed.: Prentice Hall, 2007.
using Image Processing and Neural Computing Techniques," in [9] D. Menotti, A. A. Araujo, L. Najman, G. L. Pappa, and J. Facon,
8th International Conference on Artificial Neural Networks, "Contrast Enhancement in Digital Imaging using Histogram
Skövde, Sweden, 1998, pp. 797-802. Equalization," in VII - Workshop of Theses and Dissertations /
[4] M.G. Forero, F. Sroubek, and G. Cristobal, "Identification of XXI Brazilian Symposium on Computer Graphics and Image
tuberculosis bacteria based on shape and color," Real-Time Processing, Campo Grande, Brazil, 2008, pp. 86–95.
Imaging vol. 10, pp. 251–262, 8 August 2004.
[5] S.I. Sahidan, M.Y. Mashor, A.S.W. Wahab, Z. Salleh, and H.
Ja’afar, "Local and Global Contrast Stretching For Color Contrast
Enhancement on Ziehl-Neelsen Tissue Section Slide Images " in
IFMBE Proceedings 4th Kuala Lumpur International Conference

a. c.

b. d.
Figure 6. Results of applying a. linear stretching using local minimum and maximum value b. linear stretching using average minimum and
maximum value c. histogram equalization using separate RGB histogram and d. histogram equalization using mean histogram

435