Author’s Accepted Manuscript

Lossless Image Compression Based on Integer

Discrete Tchebichef Transform

Bin Xiao, Gang Lu, Yanhong Zhang, Weisheng Li,

Guoyin Wang

www.elsevier.com/locate/neucom

PII: S0925-2312(16)30690-7
DOI: http://dx.doi.org/10.1016/j.neucom.2016.06.050
Reference: NEUCOM17302
To appear in: Neurocomputing
Received date: 10 November 2015
Revised date: 7 June 2016
Accepted date: 17 June 2016
Cite this article as: Bin Xiao, Gang Lu, Yanhong Zhang, Weisheng Li and
Guoyin Wang, Lossless Image Compression Based on Integer Discrete
Tchebichef Transform, Neurocomputing,
http://dx.doi.org/10.1016/j.neucom.2016.06.050
This is a PDF file of an unedited manuscript that has been accepted for
publication. As a service to our customers we are providing this early version of
the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting galley proof before it is published in its final citable form.
Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
 Lossless Image Compression Based on Integer Discrete Tchebichef Transform
Bin Xiao*, Gang Lu, Yanhong Zhang, Weisheng Li, Guoyin Wang
Chongqing Key Laboratory of Computational Intelligence, Chongqing University of Posts and
Telecommunications, Chongqing, 400065, China
*
Corresponding author: Tel.: +86 23 62461429. [email protected]

Abstract:

Transform coding plays a very important role in image and video compression. Discrete Cosine

Transform (DCT) is used as standard scheme (i.e. JPEG) in lossy image compression. Consequently,

integer Discrete Cosine Transform (iDCT) is presented to achieve lossless compression for the

compatibility of JPEG. Presently, with the investigation of new and well performed image transform

techniques, there is an undeniable need for novel transform coding technologies to improve the

compression rates and reduce computational complexity in the field of transform based lossless image

compression. Discrete Tchebichef Transform (DTT) is a potentially unexploited orthogonal transform,

and has shown a number of valuable properties like energy compaction and recursive computation. It

has been preliminarily introduced in lossy image compression and shown the superiority in the

compression rates. However, the DTT has not been investigated in lossless image compression. In this

paper, we study DTT and matrix factorization theory firstly, and then factorize the N×N DTT matrix

into N+1 single-row elementary reversible matrices (SERMs) with minimum rounding errors. On this

base, we introduce a novel algorithm, named integer DTT (iDTT), to achieve integer to integer

mapping for efficient lossless image compression. A series of experiments are carried out and results

show that the proposed iDTT algorithm not only has higher compression ratio than iDCT method, but

also is compatible with the widely used JPEG standard.

Keywords: Discrete Tchebichef Transform; discrete cosine transform; lossless image compression;

JPEG; matrix factorization; image transform.

1. Introduction

Image and video compression (coding) has played an important role in the application of image

transmission and storage. The scheme of image coding can be categorized into two distinct kinds,

lossless and lossy coding. Lossy coding which provides great compression gains at the expense of

information integrity has been widely used in digital camera, World Wide Web, mobile device and so
on. On the contrary, lossless coding that holds the information integrity throughout the entire encoding

and decoding process has been introduced as a necessary procedure in the application of medical

images [1], remote sensing [2][3], satellite communications [4], etc.

Transform coding techniques such as Karhunen-Loeve transform (KLT) [5], discrete cosine

transform (DCT) [6] and Discrete Wavelet transform (DWT) [7] [8] are very important tools in image

and video compression. The most well know transform coding is the DCT. Since DCT is the best

substitution of KLT which is considered to be statistically optimal for energy concentration [9], it has

been commonly used in many image compression standards such as JPEG [10], MPEG-2 [11] and

H.263 [12]. Some state-of-art technologies such as JPEG-2000 and JPEG-XR still cannot be a

substitute for JPEG due to: 1) JPEG is the most popular image compression format over the world; 2)

JPEG2000 is historically covered by patents and requiring royalty-bearing licenses.

For lossless image compression, methods can be divided into two categories: predictive coding and

transform coding. Predictive coding is one of the key techniques in the field of lossless image

compression. The well-known predictive coding algorithms are JPEG-LS [1] and CALIC [13] that use

single predictor (median edge detection predictor in JPEG-LS and gradient adaptive predictor in

CALIC) for lossless image compression. On this base, multi-predictor combination methods such as

adaptive prediction combination [14] and Bayesian predictor combination [15] methods are presented

to improve the compression rates. However, those predictive coding based algorithms are realized in

the spatial domain and independent with the scheme of JPEG. To be compatible with the scheme of

JPEG, recently, transform coding such as integer DCT (iDCT) based methods are presented for lossless

image compression. One method for developing an iDCT algorithm is based on the idea of

approximating the components of the cosine matrix with order n by dyadic rotationals [16]. This

method destroys the orthogonality of the transform matrix, and the challenge of this method is to find

an invertible approximation matrix. That is exactly the reason only order n =8 or n =16 of the

approximated cosine matrix have been provided in this method [16]. The other methods [17]-[21]

factorize the transform matrix into products of lifting matrices and simple matrices. These methods

have the advantages that keeping the orthognality of the transform matrix and working suitable for

arbitrary size of the matrix.

DCT based image compression standard (JPEG) typically belongs to transform based image

compression methods and has been introduced around 25 years. This means that it’s desirable to
investigate new transform to improve the efficiency of transform based image compression and keep

the compatibility with the scheme of JPEG. Orthogonal discrete polynomial transform such as

Tchebichef [22], Krawtchouk [23], Racah [24], dual Hahn [25] and Hahn [26] with the properties of

energy compaction, perfect image reconstruction performance and recursive computation have been

extensively employed in fields of pattern recognition [27], robot vision [28], watermarking [29], etc.

Recently, Prattipati, S. et al. [30] [31] introduced DTT for lossy image compression under the scheme

of JEPG, and their experimental results show that DTT provides a similar performance with DCT in

lossy image compression. Unfortunately, the lossless image compression using this orthogonal discrete

polynomial transform, so far, has not been investigated.

In this paper, we propose an integer DTT (iDTT) algorithm that realizes integer to integer mapping

by factorizing N  N transform matrix into N  1 single-row reversible matrices for lossless image

compression. The proposed compression framework is compatible with JPEG and can significantly

improve the image compression performance when compared with existing iDCT. To the best of our

knowledge, this is the first attempt to introduce orthogonal discrete polynomial transform in lossless

image compression.

The rest of paper is organized as follows: In Section 2, we introduce some mathematic backgrounds

such as the definition of DTT, factorization of DTT matrix to achieve iDTT and a discussion of error

preservation. The framework of lossless image compression using iDTT is given in section 3. Section 4

provides experimental validations of the developed theoretical framework. The comparisons on lossless

image compression performance for gray, color, medical images using JPEG-LS, iDCT and our

proposed iDTT are also provided in this section. Section 5 concludes this paper.

2. DTT and iDTT

2.1 DTT

DTT is built on discrete Tchebichef polynomials (DTP), and DTP is a class of hypergeometric

orthogonal polynomials. The n -th order scaled DTP is defined as follows [22]

(1)

where tn ( x) is the original DTP

nk
n
 N  1  k  n  k  x 
tn ( x)  n ! (1)     (2)
k 0  n  k  n  k 
and  (n, N ) is the squared-norm of which can be written as

N ( N 2  1)( N 2  22 ) ( N 2  n2 )  N  n
 (n, N )   (2n)!  (3)
2n  1  2n  1 
The scaled DTP satisfies the following orthogonality condition

(4)

Moreover, the scaled DTP has the following recurrence relation

(a1 x  a2 )tn 1 ( x)  a3tn  2 ( x)

tn ( x)  (5)
n

1 2x  1  N 3
with t0 ( x)  , t1 ( x)  and
N N N ( N 2  1)

2 4m 2  1 1 N 4m 2  1
a1  , a2 
m N m
2 2
m N 2  m2
(6)
1  m 2m  1 N 2  (m  1) 2
a3 
m 2m  3 N 2  m2

For a N  M 2-D input signal (image) X  X  i, j  , the transformed matrix Y  Y (n, m) after forward

DTT can be written as

N 1 M 1
Y (n, m)    tn (i)tm ( j ) X (i, j ) (7)
i 0 j 0

Consequently, the inverse DTT is defined as

N 1 M 1
X (i, j )    Y (n, m)tn (i)tm ( j ) (8)
n 0 m0

with i, n  0,1, 2, , N 1 and j, m  0,1, 2, , M 1 .

Similarly, the forward DTT and its inverse version defined in Eqs. (7) and (8) can also be written in

form of matrix as

Y  A  X  AT and X  AT  Y  A (9)

where AT  A1 and .

2.2 Integer DTT (iDTT)

As proved in [32], for a N  N nonsingular matrix A , if det(A) =1, it can be factorized into N +1

single-row elementary reversible matrices (SERMs)

A  PSN S2 S1S0 (10)

where P is the permutation matrix and S k with k  [0, N ] are the k-th SERMs. Consequently, according
to the definition of DTT in Eqs. (1) and (9), the coefficients of 8×8 DTT transform matrix A is

 0.3536 0.3536 0.3536 0.3536 0.3536 0.3536 0.3536 0.3536 

 -0.5401 0.3858 0.2315 0.0772 0.0772 0.2315 0.3858 0.5401

 0.5401 0.0772 0.2315 0.3858 0.3858 0.2315 0.0772 0.5401
 
-0.4308 0.3077 0.4308 0.1846 -0.1846 0.4308 0.3077 0.4308 
A
 0.2820 0.5238 0.1209 0.3626 0.3626 0.1209 0.5238 0.2820 
 
 0.1498 0.4922 0.3638 0.3210 0.3210 0.3638 0.4922 0.1498 
 0.0615 0.3077 0.5539 0.3077 0.3077 0.5539 0.3077 0.0615 
 
 0.0171 0.1195 0.3585 0.5974 0.5974 0.358 5 0.1195 0.0171

It’s easy to find that this 8×8 DTT transform matrix satisfies AT  A1 and det(A) =1. Therefore, it can

be factorized into 9 SERMs using (10) with

Sm  I  em smT (m  1, 2 ,8) (11)

and S0  I  e8 s0T . Where em is the m-th column of 8×8 identity matrix, and smT are the vectors to

make SERMs defined in Eq. (10) with the m-th ( m  1, 2 ,7 ) element is zero

sT0  -0.8516 1.4771 1.9555 1.2229 2.6259 0.2772 5.2787 0 

 T 
s1   0 0.8749 0.8247 0.2747 -1.8039 -0.3812 2.7737 0.5401 
sT2  -0.0639 0 1.3260 0.7290 1.4312 0.5746 -2.7592 0.4653 
 T  
s3   0.0937 0.3165 0 0.0330 0.7168 0.3984 1.1899 0.1473 
sT    0.4161 0.3273 0.7480 0 0.2929 0.6221 1.2305 0.1523 
 T4   
s5   0.5027 0.4629 0.1874 0.3886 0 0.8798 1.7402 0.2154 
sT   0.1659 0.9679 0.0743 0.5311 0.6921 0 0.4690 0.1330 
 6  
sT7   0.0315 1.1460 0.9335 0.0854 0.2858 1.7895 0 0.4922 
 T  
s8   0.2207 2.9537 2.5221 1.2378 1.2992 4.4198 2.2747 2 

The corresponding permutation matrix in Eq. (10) is

é 0 0 0 0 1 0 0 0 ù
ê ú
ê 0 0 0 0 0 1 0 0 ú
ê 1 0 0 0 0 0 0 0 ú
ê ú
P =ê 0 1 0 0 0 0 0 0 ú
ê 0 0 0 1 0 0 0 0 ú
ê 0 0 0 0 0 0 0 1 ú
ê ú
ê 0 0 1 0 0 0 0 0 ú
êë 0 0 0 0 0 0 1 0 úû

Therefore, the one dimensional forward iDTT for a vector X can be achieved by

Y  iDTTforward ( X )  P  S8  S1  S0 X    (12)

where  denotes the arithmetic rounding operator, X  ( x0 , x1 , xN 1 )T is the input vector, and Y is the

output vector. The inverse iDTT can be easy deduced from Eqs. (9) and (12).

X  iDTTinverse (Y )   S01  S71  S81 PT Y   

   (13)


with Sm1  I  em smT , (m  1, 2, 8) and S01  S0  I  e8 s0T .

According to Eq. (9), the forward iDTT for a 2-D input signal can be written as
 Y  iDTT forward iDTT forward ( X) 
T
 (14)

and the 2-D inverse iDTT is


X  iDTTinverse  iDTTinverse (Y) 
T
 (15)

Due to the rounding arithmetic applied in each SERM, floating-point values storage and calculation

with fixed-precision computers [39], some errors will definitely be involved. For the SERMs

factorization used in (10), the total rounding error vector u results from the transform of m-th
SERMs is
u  P  (u1  S N  (u2  S N 1 (uN  S1  uN 1 ) ))
N 1 m 1 (16)
 P  u1  P   ( S N  k 1 )  um
m2 k 1

If all the elements in the row of S m are integers and um is equal to 0, the transform matrix will not

produces any errors, otherwise, the previous rounding off errors will be transferred step by step:
1 m 1 1
| u | u  P {  (|  S
m  N 1 k N
k |)  em  ( I  |  Sk |)  en }
k N
(17)

where |  | denotes the absolute value in the vector. A factorization with partial or complete pivoting is

necessary in order to minimize errors. However, the row-exchange criterion still deserves further

investigation since the permutation matrix P mentioned in Section 2.2 is not unique. We use minimum

| Sm | for as a criterion to make sure the error produced in each step is less than 0.5 in the

proposed algorithm. Consequently, the lossless integer to integer mapping can be realized.

3. Lossless image compression using iDTT

3.1 The scheme of iDTT for lossless image compression

The framework of proposed lossless image compression and decompression scheme is compatible

with DCT based compression scheme-JPEG. The improvements of proposed scheme rely on the

following aspects:

 forward iDTT replaces the original forward float DCT in the encoder of JPEG;

 inverse iDTT takes place with the inverse float DCT in the decoder of JPEG;

 the elements of quantization and de-quantization table are all 1s in both encoder and

decoder of proposed scheme.

Due to 1) the Huffman entropy and de-entropy processes are lossless; 2) the forward and inverse

iDTT are integer to integer mapping and lossless; 3) the quantization as well as de-quantization steps
with all elements are 1s do not involve any information lost in the computation of integer, the proposed

scheme is absolutely lossless. The reconstructed image is exactly the same with the original image, and

its peak signal noise ratio (PSNR) is definitely infinite.

The encoding and decoding processes of iDTT based lossless image compression are shown in Fig.1.

(a) and (b) respectively. In the proposed scheme, Huffman coding is used in both the entropy and

de-entropy steps. It is worth to note that the proposed scheme also allows both encoding and decoding

processes to support arithmetic coding, which is mathematically superior to Huffman coding.

Arithmetic coding typically makes files about 5-7% smaller than Huffman coding. However, arithmetic

coding is slower in encoding and decoding when compared to Huffman coding.

The reason we set the elements of quantization table are all 1s in both encoding and decoding

processes in our proposed scheme is that: 1) quantization and de-quantization table with all elements 1s

will not lead information lost in the scheme of iDTT since both forward and inverse iDTT are integer to

integer mapping; 3) the proposed scheme with quantization and de-quantization table is compatible

with lossy image compression. Lossy image compression can be easily realized by proposed scheme if

suitable quantization and de-quantization tables are involved.

Fig. 1 The framework of lossless image compression using iDTT (a) for encoding and (b) for decoding.
The Dashed boxes in both encoding and decoding are only for color image compression.
3.2 Reversible color transform by matrix factorization

For color images compression, a color transform should be involved as preprocess step before

coding. The determination of the color transform matrix in JPEG (i. e. RGB to YCbCr) is 0.2363 [35],

which is less than 1. It means that YCbCr removes some information to get lower entropy, and the

removed information can’t be recovered. Therefore, the traditional RGB to YCbCr in JPEG is a lossy
color transform. To achieve lossless color transform, recently, a lot of reversible color transforms are

proposed in [33] [34]. Since the matrix factorization method introduced in Section 2 can achieve

integer to integer mapping without information lost. On this base, a novel reversible color transform

based on matrix factorization for lossless color image compression is proposed. We adopt a modified

version of traditional RGB to YCbCr transform matrix by simply double expanding Cb and Cr

component, and replacing the value 0.1140 in Y component with 0.1721. This modified reversible color

transform can be written as:

(18)

From (18), we can find that the determinant of modified transform matrix is 1. Consequently, it can be

factorized into several invertible matrices for lifting:

(19)

By substituting the factorized matrices in Eq. (19) to Eqs. (12) and (13), the forward and inverse
lossless reversible color transform can be achieved.
4. Experimental results and analysis

Since the PSNR is infinite for lossless image compression, here, we use bit rates (BRs) [17] as a

metric for lossless image compression assessment, the lower the BRs is, the better the compression

performance. A comparison with that of predictive coding lossless compression (PEG-LS, proposed in

[1]) and integer DCT based lossless image compression (iDCT, proposed in [35]) is provided to

evaluate the performance of the proposed iDTT based method.

4.1 Lossless compression for gray level images

In this experiment, seven groups of miscellaneous gray level images [36] of size 256×256 are

selected to evaluate the performance of JPEG-LS, iDTT and iDCT. The results are shown in Table 1.

From this table, we can find that the BRs of both iDCT and iDTT are a little higher than JPEG-LS for

gray level images compression. This result shows that predictive coding scheme (JPEG-LS) has high

efficiency in lossless image compression. However, the scheme of JPEG-LS is not compatible with the
widely accepted scheme-JPEG since it doesn’t belong to transform coding. Whereas the scheme of

iDTT is exactly compatible with JPEG. Both lossy and lossless image compression can be realized

under the proposed iDTT scheme. Moreover, the BRs of iDTT are significantly less than that of iDCT.

This implies that DTT has better energy compaction and de-correlation properties than DCT. Therefore,

iDTT is more suitable for investigating as the transform kernel in both lossy and lossless compatible

image compression scheme.

Table 1. BRs of lossless compression for gray level images. From up to down, the first to the seventh
group.

G256_001

JPEG_LS 5.71 5.16 5.10 4.71 4.32 3.51 4.60 5.19 6.06 6.72 5.41 5.48
iDCT 7.06 6.47 5.72 6.11 5.66 4.91 5.62 5.86 6.98 6.69 6.70 6.58
iDTT 6.58 5.95 5.52 5.70 5.34 4.54 5.30 5.66 6.86 6.66 6.21 6.21

G256_002

JPEG_LS 0.47 2.85 2.44 2.91 3.13 3.14 6.02 4.71 6.06 3.17 6.30 6.78
iDCT 1.31 4.03 3.63 4.44 4.54 4.62 7.04 5.39 6.41 4.41 6.54 7.87
iDTT 0.98 3.65 3.25 3.95 4.02 4.02 6.92 5.18 6.26 4.05 6.50 7.69

G256_003

JPEG_LS 4.02 4.55 3.86 4.06 2.68 2.98 4.02 1.05 6.01 5.26 5.56
iDCT 5.07 5.89 5.53 5.46 4.12 4.29 5.85 3.53 6.99 6.33 6.29
iDTT 4.78 5.57 5.08 5.01 3.62 3.87 5.13 3.16 6.83 6.08 6.13

G256_004

JPEG_LS 3.65 3.69 2.16 4.52 4.55 5.48 4.12 4.32 3.48 4.94 5.00 4.55
iDCT 5.09 5.10 3.79 5.54 5.96 6.63 5.14 5.50 4.92 6.17 5.62 6.00
iDTT 4.63 4.56 3.38 5.22 5.56 6.31 4.86 5.18 4.34 5.90 5.47 5.44

G256_005

JPEG_LS 4.88 6.61 6.25 3.40 5.19 4.82 4.57 4.25 2.72 3.62 3.52 5.30
iDCT 6.40 7.00 6.96 5.22 6.15 5.44 6.00 4.02 4.92 5.17 6.13 6.77
iDTT 5.96 6.90 6.82 4.79 5.75 5.15 5.39 3.64 4.38 4.52 5.88 6.26

G256_006

JPEG_LS 5.23 3.85 5.33 5.88 5.88 5.79 4.85 5.63 5.30
 iDCT 6.04 4.76 5.87 6.51 6.51 6.32 5.44 6.20 5.99
iDTT 5.83 4.60 5.75 6.34 6.35 6.17 5.33 6.05 5.76

G256_007

JPEG_LS 5.16 6.02 5.19 5.59 4.76 5.63 5.48 4.00 4.91 6.90 7.62
iDCT 6.47 7.12 5.63 6.10 6.09 8.60 6.04 7.60 5.59 7.36 7.85
iDTT 5.95 6.89 5.60 6.04 6.05 8.03 5.98 7.45 5.48 7.25 7.79

4.2 Lossless compression for color images

24 true color images of size 1024×768 or 768×1024 available on publicly database [37] are selected

for testing. The BRs of those color images lossless compression using JPEG-LS, iDCT and iDTT are

shown in Table 2. It can be seen that, the BRs of iDTT is dramatically lower than iDCT and JPEG-LS.

It is undeniable to declare that the proposed iDCT has higher compression ratio when compared with

iDCT and JPEG-LS. The reasons that iDTT performs better than JPEG-LS and iDCT in color image

compression are: 1) the reversible color transforms is involved in the scheme of iDTT which introduces

greater compression without loss on image quality; 2) iDTT has better performance on energy

compaction and information de-correlation than iDCT.

Table 2. BRs of lossless compression for color images. From up to down, the first to the second group
Color
images

JPEG_LS 4.57 4.52 3.54 4.52 3.05 4.06 3.88 4.91 5.93 3.83 4.40 5.26
iDCT 4.66 4.67 4.18 4.06 3.78 3.97 4.17 4.82 5.12 3.84 4.32 4.18
iDTT 4.35 4.31 3.84 3.70 3.63 3.77 3.88 4.39 4.82 3.67 4.03 3.77

Color
images

JPEG_LS 3.64 4.55 5.16 3.49 4.04 5.26 4.44 5.06 4.10 3.92 3.92 4.17
iDCT 4.18 4.38 5.22 3.82 4.23 4.48 4.33 4.67 4.99 4.12 4.08 4.31
iDTT 3.77 4.15 4.64 3.57 4.03 4.30 4.08 4.31 4.56 3.85 3.83 3.96

4.3 Lossless compression for medical images

In the application of medical image storage and transmission, lossless image compression is

absolutely required due to the possible loss of useful clinic information in the lossy compression. In

this experiment, a digital image dataset of complete human male and female cadavers in MRI, CT and

anatomical modes [38] (most of these images approximately 3.5 megabytes in size, the full opened

image being about 7 megabytes in size) is used for evaluation. The compression results are shown in

Table 3. From this table we can find that the BRs of iDTT, iDCT and JPEG-LS are significant lower
than natural image used in section 4.1 and 4.2 due to the black background in medical images. In

addition, as we shown in the previous experiments, iDTT performs better in color medical image

lossless compression than iDCT and JPEG-LS. iDTT has better performance than iDCT in both gray

level and color medical images. JPEG-LS still has a little lower BRs than iDCT and iDTT in gray level

(or monochromatic) medical image compression.

Table 3. BRs of lossless compression for medical images. From up to down, the first to the third group

MRI
Images

JPEG_LS 2.96 3.83 3.98 3.54 3.31 2.43

iDCT 4.49 5.20 5.73 5.22 5.08 3.90
iDTT 4.02 4.62 5.01 4.60 4.48 3.55

CT
Images

JPEG_LS 1.33 1.63 2.66 1.92 2.35 1.77 2.23 0.81

iDCT 2.38 3.08 3.63 2.79 3.10 2.51 3.13 1.66
iDTT 2.08 2.56 3.34 2.55 2.92 2.34 2.88 1.50

Anatomic
Images

JPEG_LS 3.23 3.48 3.09 3.07 2.67 2.46 3.21 3.03

iDCT 3.57 3.66 3.19 3.43 3.14 3.13 3.40 3.13
iDTT 3.31 3.36 2.87 3.01 2.65 2.60 2.83 2.72

5. Conclusion

In this paper, we proposed a scheme of lossless image compression using iDTT transform. On the

contrary of predictor coding based scheme, the proposed scheme is compatible with widely used

compression scheme-JPEG. Both lossy and lossless image compression can be realized under the

proposed scheme if suitable quantization and de-quantization tables are involved. The performance of

the proposed algorithm is validated to be much better than that of iDCT based method for gray and

color image lossless compression. It is feasible that iDTT can replace iDCT in image as well as video

lossless compression due to its valuable properties such as energy compaction and information

de-correlation.

Acknowledgements:

This work is supported by the National Natural Science Foundation of China (61572092, U1401252)
and Chongqing Outstanding Youth Fund (cstc2014jcyjjq4000-1). The authors would like to thank the

anonymous referees for their valuable comments and suggestions.

References:
[1] M. J. Weinberger, G. Seroussi, and G. Sapiro, The LOCO-I lossless image compression algorithm:
Principles and standardization into JPEG-LS, IEEE Trans. Image Processing. 9 (8) (2000) 1309–
1324.
[2] L. Zhang, L. Zhang, D. Tao, X. Huang and B. Du, Compression of hyperspectral remote sensing
images by tensor approach. Neurocomputing, 147 (2015): 358-363.
[3] C. Shi, J. Zhang and Y. Zhang, Content-based onboard compression for remote sensing images.
Neurocomputing, 191 (2016): 330-340.
[4] X. Hou, M. Han, C. Gong and X. Qian, SAR complex image data compression based on quadtree
and zerotree coding in discrete wavelet transform domain: A comparative study. Neurocomputing,
148 (2015): 561-568.
[5] D. Zhang and S. Chen, Fast image compression using matrix K–L transform. Neurocomputing, 68
(2005): 258-266.
[6] K. R. Rao and P. Yip, Discrete Cosine Transform-Algorithm, Advantage, Applications, Academic
Pulishers, New York, 1990.
[7] X. Hou, L. Zhang, C. Gong, L. Xiao, J. Sun and X. Qian, SAR image Bayesian compressive
sensing exploiting the interscale and intrascale dependencies in directional lifting wavelet
transform domain. Neurocomputing, 133 (2014): 358-368.
[8] A.J. Hussain, D. Al-Jumeily, N. Radi and P. Lisboa, Hybrid neural network predictive-wavelet
image compression system. Neurocomputing, 151 (2015): 975-984.
[9] T. I. Haweel, A new square wave transform based on the DCT, IEEE Trans. Signal Process. 81 (11)
(2001) 2309-2319.
[10] Information Technology: Digital Compression and Coding of Continuous-Tone Still Images:
Requirements and Guidelines, document ISO/IEC 10918-1:1994, ITU-T Rec. T.81, 1992.
[11] Generic Coding of Moving Pictures and Associated Audio Information Part 2: Video, document
ISO/IEC 13818-2 MPEG-2, ITU-T Rec. H.262, 1998.
[12] Video Coding for Low Bit Rate Communication, ITU-T Rec. H.263, Jan. 2005.
[13] X. Wu, N. Memon, CALIC—A context based adaptive lossless image codec, in Proc. IEEE
ICASSP. 4(1996) 1890–1893.
[14] G. Deng, H. Ye, L. Cahill, Adaptive techniques for lossless data compression, in Proc. Austral.
New Zealand Conf. Intell. Inf. Syst., (2001) 345–350.
[15] A. Martchenko, G. Deng, Bayesian Predictor Combination for Lossless Image Compression. IEEE
Transactions on Image Processing. 22 (12) (2013) 5263-5270.
[16] T. D. Tran, The BinDCT: Fast multiplierless approximation of the DCT, IEEE Signal Process. Lett.
7 (6) (2000) 141-144.
[17] L. Z. Cheng, H. Xu, Y. Luo, Integer discrete cosine transform and its fast algorithm, Electron. Lett.
37 (1) (2001) 64-65.
[18] J. Liang, T. D. Tran, Fast multiplierless approximations of the DCT with the lifting scheme, IEEE
Trans. Signal Process. 49 (12) (2001) 3032-3044.
[19] Y. Zeng, L. Cheng, G. Bi, A. C. Kot, Integer DCTs and fast algorithms, IEEE Trans. Signal
Process. 49 (11) (2001) 2774-2782.
[20] T. Suzuki, M. Ikehara, Integer DCT Based on Direct-Lifting of DCT-IDCT for Lossless-to-Lossy
Image Coding, IEEE Trans. Image Process., 19 (11) (2010) 2958-2965.
[21] G. Plonka, M. Tasche, Invertible integer DCT algorithms, Applied and Computational Harmonic
Analysis, 15 (1) (2003) 70-88.
[22] R. Mukundan, S. H. Ong, P. A. Lee, Image analysis by Tchebichef moments, IEEE Trans. Image
Process. 10 (9) (2001) 1357-1364.
[23] P.T. Yap, R. Paramesran, O. Seng-Huat, Image analysis by Krawtchouk moments, IEEE Trans.
Image Process. 12 (11) (2003) 1367- 1377.
[24] H. Zhu, H. Shu, J. Liang, L. Luo, J. Coatrieux, Image analysis by discrete orthogonal Racah
moments, IEEE Trans. Signal Process. 87 (4) (2007) 687-708.
[25] H. Zhu, H. Shu, J. Liang, L. Luo, J. Coatrieux, Image analysis by discrete orthogonal dual Hahn
moments, Pattern Recognit. Lett. 28 (13) (2007) 1688-1704.
[26] Y. Pew-Thian, R. Paramesran, O. Seng-Huat, Image Analysis Using Hahn Moments, IEEE Trans.
Pattern Anal. Machine Intell. 29 (11) (2007) 2057-2062.
[27] E.G. Karakasis, G. A. Papakostas, D. E. Koulouriotis, V. D. Tourassis, Generalized dual Hahn
moment invariants, Pattern Recognit. 46 (7) (2013) 1998-2014.
[28] S. P. Priyal, P. K. Bora, A robust static hand gesture recognition system using geometry based
normalizations and Krawtchouk moments, Pattern Recognit. 46 (8) (2013) 2202-2219.
[29] L. Zhang, G. B. Qian, W. W. Xiao, Z. Ji, Geometric invariant blind image watermarking by
invariant Tchebichef moments, Opt. Express, 15 (5) (2007) 2251-2261.
[30] S. Prattipati, S. Ishwar, M. N. S. Swamy, P. K. Meher, A fast 8×8 integer Tchebichef transform
and comparison with integer cosine transform for image compression, in: Circuits and Systems
(MWSCAS), IEEE 56th International Midwest Symposium. 8 (2013) 1294–1297.
[31] S. Prattipati, M.N.S. Swamy, P.K. Meher, A variable quantization technique for image
compression using integer Tchebichef transform, in: Information, Commun. and Signal Process.
(ICICS) 9th International Conference. 11 (2013) 1-5.
[32] P. W. Hao, Q. Y. Shi, Matrix factorizations for reversible integer mapping, IEEE Trans. Signal
Process. 49 (10) (2001) 2314-2324.
[33] T. Strutz, Multiplierless Reversible Color Transforms and Their Automatic Selection for Image
Data Compression, IEEE Transactions on Circuits and Systems for Video Technology. 23 (7)
(2013) 1249-1259.
[34] M. Iwahashi, H. Kiya, Reversible color transform with compatibility to irreversible transform,
17th IEEE International Conference on Image Processing (ICIP). (2010) 2881 - 2884
[35] Y. Chen, P. W. Hao, Integer reversible transformation to make JPEG lossless, in: IEEE 7th
International Conference on ICSP Proceeding. 1 (2004) 835-838.
[36] CVG-UGR image database. [Online]. Available: http://decsai.ugr.es/cvg/dbimagenes/g256.php
[37] Kodak: 24 PNG images. [Online]. Available: http://r0k.us/graphics/kodak
[38] Medical images. [Online]. Available: http://www.nlm.nih.gov/research/visible/getting_data.html
[39] P. W. Hao, Q. Y. Shi, Invertible linear transforms implemented by integer mapping, Science in
China, 30 (2) (2000) 132-140.

Author biography
Bin Xiao was born in 1982. He received his B.S. and M.S. degrees in Electrical Engineering from
Shaanxi Normal University, Xi'an, China, in 2004 and 2007, received his Ph.D. degree in computer
science from Xidian University, Xi'An, China in 2012. He is now working as an associate professor at
Chongqing University of Posts and Telecommunications, Chongqing, China. His research interests
include image processing, pattern recognition and digital watermarking.

Gang Lu was born in 1989. He received the B.S. degree in Computer science and technology from
Gansu Institute of Political Science and Law, China in 2013. He is currently pursuing the M.S. degree at
Chongqing University of Post and Telecommunications. His research interests include feature
extraction and image compression.

Yanhong Zhang was born in 1991. He received the B.S. degree in Computer science and technology
from Henan University of Economics and Law, China in 2014. She is currently pursuing the M.S.
degree at Chongqing University of Post and Telecommunications. Her research interests include image
analysis and pattern recognition.

Weisheng Li received his B.S. degree from School of Electronics and Mechanical Engineering at Xidian
University, Xi'an, China, in July 1997. He received his M.S. and Ph.D. degrees from School of Electronics
and Mechanical Engineering and School of Computer Science and Technology at Xidian University, in
July 2000 and July 2004, respectively. Currently he is a professor of Chongqing University of Posts and
Telecommunications. His research focuses on intelligent information processing and pattern
recognition.

Guoyin Wang was born in 1970. He received the B.S. degree in computer software, the M.S. degree
in computer software, and the Ph.D. degree in computer organization and architecture from Xi'an
Jiaotong University, Xi'an, China, in 1992, 1994, and 1996, respectively. He is the author of over 200
reviewed research publications. His books and papers have been cited over 6000 times. His research
interests include data mining, machine learning and pattern recognition.

Bin Xiao
 Gang Lu

Yanhong Zhang

Weisheng Li

Guoyin Wang

Highlights
 A framework of lossless image compression based on integer DTT is proposed.
 The compression efficiency of integer DTT is higher than integer DCT.
 The proposed lossless image compression scheme is compatible with JPEG
 Both lossy and lossless image compression can be realized under the proposed scheme
 A lossless color transform method based on traditional RGB to YCbCr is presented.