Adnan Khashman and Kamil Dimililer
Department of Electrical & Electronic Engineering, Near East University, Lefkosa, Northern Cyprus
Keywords: Intelligent Systems, Neural Network, DCT Image Compression, Optimum Compression.
Abstract: The parallel processing capability of neural networks provides efficient means for processing images with
large amount of data. Image compression using Discrete Cosine Transforms (DCT) is a lossy compression
method where at higher compression ratios the quality of the compressed images is reduced, thus the need
for finding an optimum compression ratio that combines high compression and good quality. This paper
suggests that the image intensity can affect the choice of an optimum compression ratio. A neural network
will be trained to establish the non-linear relationship between the image intensity and its compression ratios
in search for an optimum ratio. Experimental results suggest that a trained neural network can relate image
intensity or pixel values to its compression ratio and thus can be successfully used to predict optimum DCT
compression ratios for different images.
Image compression is one of these commonly used
image processing applications where the rapid
advance in multimedia applications has made data
compression more vital and compression methods
are being developed to compress large data files
such as images (Nadenau etal., 2003). Efficient
methods usually succeed in compressing images,
while retaining high image quality and marginal
reduction in image size (Ratakonda and Ahuja,
Recently, adaptive prediction was suggested for
data compression and still image coding (Robinson,
2006), and oriented wavelet transform for image
compression was proposed (Chappelier and
Guillemot, 2006). The use of wavelet transforms and
Discrete Cosine Transform (DCT), when applied to
image compression, was also investigated
(Khashman and Dimililer, 2005). The usability and
efficiency of these methods depend on the
application areas that require either high
transmission rate or high quality decompression.
Lossless compression algorithm provides a
compression which, when decompressed the exact
original data can be obtained. This is the case when
binary data such as executables and documents are
compressed. On the other hand, images might not be
reproduced 'exactly', but an approximation of the
original image is enough for most purposes as long
as the error between the original and the compressed
image is tolerable. The general purpose of
compression programs is to compress images, but
the result is less than optimal.
DCT-based image compression is a simple
compression method that was first applied in 1974
(Ahmed etal., 1974). The disadvantage of using
DCT image compression is the high loss of quality
in compressed images, which is more notable at
higher compression ratios. Recent work on finding
optimum compression suggested a criterion based on
visual inspection and computed analysis of the
reconstructed images (Khashman and Dimililer,
2005). Visual inspection and observation by humans
is an empirical analysis that involves a number of
people who observe the smoothness and edge
continuity of certain objects within reconstructed
images and then decide which compression ratio
provides a compromise between high compression
ratio and minimal loss of quality (Jahne, 2002),
(Khashman and Dimililer, 2005), i.e. the optimum
compression ratio.
The use of neural networks for image processing
applications has marginally increased in recent
years, where image compression using DCT and a
neural network was suggested previously (Ng and
Cheng, 1997). More recently, different image
compression techniques were combined with neural
network classifiers for various applications (Zhou
etal., 2006), (Milani, 2006). However, none of these
methods suggested using a neural network to
Khashman A. and Dimililer K. (2007).
In Proceedings of the Ninth International Conference on Enterprise Information Systems - AIDSS, pages 451-454
DOI: 10.5220/0002402204510454
determine optimum compression using the original
image intensity.
The aim of the work presented within this paper
is to develop an intelligent system for optimum DCT
image compression using a neural network. Our
hypothesis is that a trained neural network can learn
the non-linear relationship between the image
intensity (pixel values) and its optimum compression
ratio. Once trained, the neural network would predict
the optimum compression ratio of an image upon
presenting the image to the neural network by using
its intensity values. The prediction parameters here
are the image intensity (global pixel values) and the
experience (training) of the neural network.
The development and implementation of the
proposed intelligent system for optimum DCT image
compression uses 80 images that have different
objects, brightness and contrast. The novel system is
implemented in two phases: the image processing
phase and the neural network arbitration phase.
2.1 Image Pre-Processing Phase
DCT-based image compression is firstly applied to
all 80 images using 9 compression ratios (10%,
20%,… 90%) as shown in an example in Figure 1.
The optimum DCT compression ratios for the 80
images are then determined using the optimum
compression criteria based on visual inspection as
suggested by (Khashman and Dimililer, 2005).
The image database is then organized as follows:
40 images in the training image set and 40 images in
the testing image set which will be used to verify the
efficiency of the proposed method. Figure 2 shows
the examples of original images and their
compressed versions using optimum compression
ratios prior to training the neural network.
2.2 Neural Network Phase
The intelligent image compression system uses a
supervised neural network based on the back
propagation learning algorithm due to its
implementation simplicity, and the availability of
sufficient database for training this supervised
learner. The neural network consist of an input layer
with 4096 neurons, one hidden layer with 82
neurons and an output layer with 9 neurons. Training
the neural network uses 40 images which are grey
and of size (256x256) pixels. Using Adobe
Photoshop, the size of each image is initially
reduced to (64x64) pixels prior to presenting the
whole reduced image to the neural network, thus
resulting in 4096 pixel values per image. Further
reduction to the size of the images was attempted in
order to reduce the number of input layer neurons
and consequently the training time, however,
meaningful neural network training could not be
achieved, thus the use of whole images of size
(64x64) pixels. The hidden layer of the neural
network contains 82 neurons and he output layer has
nine neurons according to the number of possible
compression ratios (10% - 90%). Figure 3 shows the
topology of this neural network where ODCR stands
for optimum DCT Compression Ratio.
The successful implementation of the proposed
intelligent image compression system relies mainly
on the learning capability of the neural network
within the intelligent system. Meaningful learning
and correct association of original images to their
optimum compression ratios relies on the provision
of sufficient input data patterns to the neural
network during training and later on during
3.1 Neural Network Performance
The neural network learnt after 4447 iterations and
within 3300 seconds. The running time for the
generalized neural network after training and using
Original Image 10% 20% 30% 40%
50% 60% 70% 80% 90%
Figure 1: An image and its compression at nine ratios.
a- b- (30%) a- b- (50%)
Figure 2: (a) Original images and (b) their optimum
ICEIS 2007 - International Conference on Enterprise Information Systems
Table 1: Neural Network Final Training Parameters.
Input nodes 4096
Hidden nodes 82
Output nodes 9
Learning rate
Momentum rate
Iterations 4447
Training time (seconds) 3300
Run time (seconds)
Table 2: Accuracy and recognition rates for OCD.
Accuracy Rate
Recognition Rate
0 100 % 12/40 (30 %)
1 89 % 31/40 (78 %)
2 78 % 40/40 (100 %)
one forward pass was 0.003 seconds. These results
were obtained using a 2.0 GHz PC with 2 GB of
RAM, Windows XP OS and Matlab 7.1 software.
Table 1 list the final parameters of the successfully
trained neural network.
3.2 Evaluation Method
The evaluation of the training and testing results was
performed using two measurements: the recognition
rate and the accuracy rate. The recognition rate is
defined as follows:
where RR
is the recognition rate for the neural
network within the intelligent system, I
is the
number of optimally compressed images, and I
the total number of images in the database set.
The accuracy rate RA
for the neural network
output results is defined as follows:
where S
represents the pre-determined (expected)
optimum compression ratio, S
represents the
optimum compression ratio as determined by the
trained neural network and S
represents the total
number of compression ratios.
Table 2 shows the three considered OCD values
and their corresponding accuracy rates and
recognition rates.
The Optimum Compression Deviation (OCD) is
another term that is used in our evaluation. OCD is
the difference between the the pre-determined or
expected optimum compression ratio (S
) and the
optimum compression ratio (S
) as determined by the
trained neural network, and is defined as follows:
The OCD is used to indicate the accuracy of the
system, and depending on its value the recognition
rates vary.
The evaluation of the intelligent system
implementation results in this work uses (OCD = 1)
and (OCD = 2) as they assure accuracy rates of 89%
and 78% respectively, which is considered sufficient
for this application. The trained neural network
recognized correctly the optimum compression
ratios for all 40 images in the training set as would
be expected, thus yielding 100% recognition of the
training set. Testing the trained neural network using
40 images that were not presented to the network
before, yielded 78% recognition rate with 89%
Original Image
(256x256) pixels
Input Hidden Output
er La
er La
Reduced Size
Image (64x64)
Figure 3: The intelligent optimum image compression system (ODCR: Optimum DCT Compression Ratio).
Compressed Image
Original 20% Original 40% Original 50% Original 30%
Image Compression Image Compression Image Compression Image Compression
Figure 4: Examples of Optimum DCT compression results obtained using the trained neural network.
accuracy rate (using OCD=1), and 100% recognition
rate with 78% accuracy rate (using OCD=2). Figure
4 shows examples of the optimally compressed
images as determined by the trained neural network.
This paper proposed a novel method for optimum
image compression that uses DCT compression and
a neural network. The method suggests that a trained
supervised neural network can learn the non-linear
relationship between the intensity (pixel values) of
an image and its optimum compression ratio, and
thus can predict the optimum DCT compression
ratio of an image upon presenting the original image
to the trained neural network. The implementation of
the proposed method uses lossy DCT image
compression where the quality of the compressed
images degrades at higher compression ratios. The
aim of an optimum ratio is to combine high
compression ratio with quality compressed image.
The proposed intelligent system that is presented
within this paper was implemented using 80 images
of various objects, contrasts and intensities. The
neural network within the intelligent system learnt to
associate 40 images with their different optimum
compression ratios within 3300 seconds. Once
trained, the neural could predict the optimum
compression ratio of an image within 0.003 seconds
upon presenting the image to the network.
The trained system uses three minimum
accuracy levels which are determined depending on
the application. In this work, minimum accuracy
levels of 78% and 89% were used, where; 100% and
78% recognition rates of correct optimum
compression ratio were obtained, respectively. This
successful implementation of our proposed method
was shown throughout the high recognition rate and
the minimal time cost when running the trained
neural network.
Future work will include the development of an
intelligent optimum image compression system
using Haar and Biorthogonal wavelet transform
compressions which produce higher quality
compressed images. Additionally, the intelligent
system development will use larger image database.
Ahmed, N., Natarajan, T., Rao, K.R., 1974. On Image
Processing and a Discrete Cosine Transform. In IEEE
Trans. Computers, C-23(1), 90-93.
Chappelier, V., Guillemot, C., 2006. Oriented Wavelet
Transform for Image Compression and Denoising. In
IEEE Trans. Image Processing, 15(10), 2892-2903.
Jahne, B., 2002. Digital Image Processing. New York:
Khashman, A., Dimililer, K., 2005. Image Compression
Using Biorthogonal Wavelet & Discrete Cosine
Transforms. In EMO Bilim Journal of the Chamber of
Electrical Engineers, Northern Cyprus, 5(16), 37-39.
Khashman, A., Dimililer, K., 2005. Comparison Criteria
for Optimum Image Compression. In IEEE
International Conference EUROCON’05, 935-938.
Milani, F.M., 2006. A Co-evolutionary Competitive
Multi-expert Approach to Image Compression with
Neural Networks. In IEEE International Conference
Engineering of Intelligent Systems, ICEIS’2006, 1-5.
Nadenau, M.J., Reichel, J., Kunt, K., 2003. Wavelet Based
Color Image Compression: Exploiting the Contrast
Sensitivity Function. In IEEE Trans. Image
Processing, 12(1), 58-70.
Ng, K.S., Cheng, L.M., 1997. Artificial Neural Network
for Discrete Cosine Transform and Image
Compression. In 4th International Conference
ICDAR’97, 675-678.
Ratakonda, K., Ahuja, N., 2002. Loseless Image
Comrpession with Multiscale Segmentation. In IEEE
Trans. Image Processing, 11(11), 1228-1237.
Robinson, J.A., 2006. Adaptive Prediction Trees for Image
Compression. In IEEE Trans. Image Processing,
15(8), 2131-2145.
Zhou, Y., Zhang, C., Zhang, Z., 2006. Improved Variance-
Based Fractal Image Compression Using Neural
Networks. In Lecture Notes in Computer Science, Vol.
3972, Springer-Verlag, Berlin Heidelberg New York,
ICEIS 2007 - International Conference on Enterprise Information Systems