2005
41
GVIP Journal, Volume 5, Issue 9, Dec. 2005
reversible integer wavelet transforms [2, 6, 10] can be meet the real-time delivery requirement [8] , because
developed using lifting scheme, have grown in they usually add a large amount of processing overhead.
popularity for lossless image compression. The Because of this the compression ratio achieved is also
compression systems [1], comprising a wavelet limited to 1.5 to 2. To reduce the amount of
transform followed by SPIHT (set partitioning in processing overhead in this paper selective bit
hierarchical trees) is an embedded image compression scrambling method is applied. In addition to this in the
algorithm with coding efficiency comparable to proposed approach to make the computation easier
EBCOT, which is employed in JPEG 2000. The lower integer wavelet transform is used. The digital image
complexity and computation requirement of SPIHT is will be scrambled before it is distributed and only
particularly attractive and also it is a successful method authorized users who have the proper key for
in the field of image compression. In this paper new decryption can access the clear content. This technique
region based modified SPIHT coding scheme together does not introduce additional bits which in turn reduce
with new modified runlength coding is used to encode the compression ratio.
the subband coefficients. Modified runlength coding
(MRLE) increases the compression ratios of the Hence in this paper, a secured lossless image
proposed approach. compression framework is proposed in which image
data are scrambled without significantly affecting the
Lossless image compression coder, such as context compression efficiency. The rest of paper is organized
based predictive coding (CALIC) [17] and universal as follows. Overview of the proposed framework is
context modeling based coding[14], the lack of a discussed in section II, Experiments and Results are
progressive transmission capability makes them presented in section III , the paper is concluded in
unsuitable for interactive applications like telematics section IV.
over large networks. Embedded coding permits the
progressive transmission of the compressed data, 2. Overview of the Proposed Framework
starting with an economical initial transmission of a Secured Lossless Image Compression can be obtained
rough image followed by gradual transmission of the by Integer wavelet transform followed by region based
refinement details, needed for the final reconstruction. modified SPIHT algorithm and selective bit scrambling.
Embedded progressive coding becomes quite important Scrambling is a disordered mixture of data. A digital
in terms of compressing medical images, because an image scrambling scheme should have a relatively
image may be stored without any loss. simple implementation, low-cost and low-delay
operation for real-time interactive applications. It
In addition to compression, secured transmission should have a minimum adverse impact on the
helps in accurate diagnosis and medical research. compressibility of the image. Digital image scrambling
Changes in data bits by the intruder make accurate method aims to meet the objectives outlined earlier.
diagnosis a difficult task. Medical researchers will be Fig1. shows a general architecture of the proposed
reluctant to provide their valuable contents if they are system. The Secured image compression and
not assured that their contents are securely protected. reconstruction addressed in this paper involves the
Hence several technologies have been developed for following steps.
information protection [7]. One is through encryption.
Traditional cryptographic algorithms such as Digital Step 1: The given image is decomposed using Integer
Encryption Standard (DES) [3] are not fast enough to wavelet transform. Integer wavelet transform can be
developed using lifting scheme.
42
GVIP Journal, Volume 5, Issue 9, Dec. 2005
Step 2: Region based modified SPIHT coding technique prediction and update steps. At the conclusion of these
is applied to the data obtained in the previous step. In lifting steps, the even and odd samples have been
the implementation of region based modified SPIHT the replaced by high pass (1) and low pass (2) coefficients
significant information is stored in three ordered lists as respectively.
in SPIHT – list of insignificant sets (LIS), List of
insignificant pixels (LIP) and list of significant pixels
(LSP). The algorithm consists of 2 main passes to
encode an image - Sorting pass and Refinement pass.
The LIS and LIP entries are encoded in the
sorting pass while the LSP entries are
encoded in the refinement pass.
Step 3: Scramble the encoded bits based on the key to
encrypt an image data.
Step 4: Modified runlength coding of the scrambled
coefficients. Transmit the output via transmission line.
Receiver applied the reverse procedure, obtained a d( i )[n] = d( i-1 )[n] – ( ∑ p( i ) [k] s( i-1 )[n-k] )+1/2 (1)
sequence of bits k
Step 5: The encoded coefficients are reformed by
descrambling the bits using the same key . s( i )[n] = s( i-1 )[n] – ( ∑u( i ) [k] d( i )[n-k] ) + 1/2 (2)
Step 6: The Inverse Region based modified SPIHT k
method is applied to find the positions of the decoded
symbols and decompress the wavelet coefficients of It should be noted that although integers are
the significant map. Go to step 2, run the algorithm transformed to integers, the coefficients p(i) [k] and u(i) [k]
sequentially and stops whenever a target rate is met are not necessarily integers. Thus, computing the integer
exactly. transform coefficients requires floating point operations.
Step 7: Inverse Integer wavelet transform is applied to More detailed information on lifting can be found in
the decompressed data obtained in the previous step to [5,14,15]. In this work, the following reversible integer
reconstruct the original image. The above said steps wavelet transforms of [5] based on the lifting scheme have
are explained in the forthcoming sections. been used with the notation (M,N) where M and N
represent the number of vanishing moments of the
analysis and synthesis high pass filters, respectively and
Lifting and Reversible Integer Wavelet Transforms are described in the remainder of this section. In each case
In most cases, the wavelet transform produces x[n] refers to the original signal and s[n] refer to the low
1
floating-point coefficients and although this allows pass coefficient and, d[n] and d [n] refer to the high pass
perfect reconstruction of the original image in theory, coefficients.
the use of finite-precision arithmetic, together with
quantization, results in a lossy scheme. Recently, new a . The (2,2) Transform
Integer wavelet transform, i.e. wavelet transforms that The (2, 2) is an interpolating transform designed to
transform integer to integer have been introduced [5, 12, have two vanishing moments for both the analyzing and
13]. In [5], it was shown that an integer version of every synthesizing high pass filters. It is defined by the
wavelet transform with finite filters can be obtained following lifting steps:
using the lifting scheme of [14]. Lifting allows a
wavelet transform to be computed quickly through a d[n]=x[2n+1]- ½ (x[2n]+x[2n+2] ) + 1/2 (3)
series of individual lifting steps and fully in place
calculation. Integer wavelet transform is used to reduce s[n]= x[2n] + ¼ (d[n-1] + d[n] ) + 1/2 (4)
the memory demands of the compression algorithm as
integers are used instead of real numbers. b . The (4,2) Transform
Fig.2 depicts the forward wavelet transform using The (4,2) is an interpolating transform designed to have
lifting. For the forward transform, the data set x is four vanishing moments for the analyzing high pass
split into two disjoint subsets. There is no restriction on filters and two vanishing moments for synthesizing high
how the data should be split, nor on the relative size of pass filters. It is defined by the following lifting steps:
each of the subsets. The only thing we need is some
procedure to join the subsets back into the original data d[n]= x[2n+1] 9/16 (x[2n]+x[2n+2] ) 1/16(x[2n – 2]
set x. The easiest possibility is split the data x into even + x[2n+4] )+ 1/2 (5)
and odd samples. Next is a series of dual lifting and
primal lifting steps, alternatively referred to as s[n]= x[2n] + ¼ (d[n -1] + d[n] ) +1/2 (6)
43
GVIP Journal, Volume 5, Issue 9, Dec. 2005
44
GVIP Journal, Volume 5, Issue 9, Dec. 2005
3. Refinement pass: occupies just a byte. Whereas in RLE [19] even a single
For each entry in the LSP significant for higher n, send occurrence of a character is replaced by 3 byte (count,
the nth most significant bit to the decoder. char) representation which is not efficient for a file
4. Decrement n by one and return to step 2 until the having more non repeated character. When MRLE is
specified bit rate is reached. In SPIHT, during encoding applied it gives better results for all monochrome
each significant coefficients sign in the set is checked images and as well as some gray scale images
for transmission, whereas in Region based Modified compared to arithmetic coding. In decoding, the
SPIHT, only the region of the set is checked. So in symbols are decoded sequentially by applying the
Region based Modified SPIHT the number of reverse procedure. The receiver with a proper key can
comparisons are reduced and also computational descramble the bits is given in the next step.
complexity is slightly reduced. Scrambling of the
encoded bits to ensure security is discussed in the next Descrambling the encoded coefficients
step. The encoded coefficients can be descrambled by the
user with the same key. Without this key, it is difficult
Selective bit scrambling of encoded coefficients to reconstruct the decoded symbols and also it is hard to
Several scrambling systems [5, 18] rely on methods find the position of the decoded symbols.
of directly distorting the visual image data such that,
without descrambling, the image appears unintelligible Inverse region based modified SPIHT decoding
to the viewer. The proposed approach, Scramble the method
encoded bits based on the key to encrypt an image. To At the decoder end it is possible to retrace the coding
increase the level of security block shuffling and block path and retrieve the significance map transmitted.
rotation is used, discussed in [15]. For the encryption of Based on this, appropriate values can be allocated to
the bits, if a codebreaker is to recover the encoded bits, each pixel position, noting that the centroid of the
an exhaustive search is required. Selective bit significance range( ׀2 n+1 - 2 n ׀/ 2 ) is added to every
scrambling approach, scrambles selected bits of significant pixel. In this manner an approximation of the
encoded coefficient obtained in the previous step to subband-decomposed image is reconstructed. The
encrypt an image data. Applying new modified coding procedure to obtain the original image is explained in
improves the compression performance and also it the next step
increases the security level, is explained in the next
step. Inverse Integer Wavelet Transform
Lifting allows simple inverse transform, of the same
Modified Runlength coding and decoding complexity as the forward one – Reversible integer
Modified runlength coding algorithm has the following wavelet transform being composed of the elementary
form. operations of the forward one, taken in reversed order.
Algorithm 1 Inverse integer wavelet transform is applied to the
0) Read the output file of SPIHT algorithm 8 bits at a decompressed data to get the integer data of the image.
time and convert it into character and store the Experimental results are given in the next section.
character in result file.
1) Compute threshold values. (given below) 3. Experiments and Results
2) Choose a separator character such that it is not The proposed secured Lossless Compression
present in the result file. algorithm is applied to various images of size 128 ×
3) Scan the characters of the result file one by one and 128 which is digitized to 8 bits/pixel is shown in Figures
count the repeated occurrence of the same character. 3(a),4(a),5(a). Figures (b) and (c) are decomposed and
i) if count < threshold , print character in the reconstructed images. This algorithm is tested for
output file various images and the CR (Compression Ratio ) of
ii) if count >= threshold , print the separator other lossless compression techniques such as Low
followed by count and that character. complexity, content based lossless compression
MRLE representation takes 4 bytes(separator (S)-1byte, (LOCO-I) [13], Context based image adaptive lossless
Number of characters (Count)-2, Character-1). image codec (CALIC) [17] ,EZW and arithmetic coding
Threshold = S+ Count+ Character ,SPIHT and arithmetic coding, SPIHT and Modified
runlength coding are compared and shown in Table 1.
So more than four consecutive occurrence of the same These results show that the proposed algorithm offers
character is replaced by 4 bytes representation of excellent compression ratio and also high computational
MRLE. If the number of consecutive occurrence is less speed. After analyzing the decomposition results of
than 4 the MRLE is not applied and each character various wavelet transforms, in most of the transforms all
the coefficient in the first quadrant are found to be
45
GVIP Journal, Volume 5, Issue 9, Dec. 2005
positive. So in region based modified SPIHT, the first quadrant are found to be positive. The average PSNR
quadrant is always assumed to be positive. Hence the for this work is roughly 3dB or 4dB better than
PSNR ratio of both SPIHT and Region based Modified CALIC[17] and LOCO-I [13]. There was no single
SPIHT are same with a significant increase of transform that performed best over the entire data set.
compression ratio by 0.4 to 0.6 in Region based Compared to arithmetic coding, proposed MRLE
Modified SPIHT and it is shown in Table1. But in some algorithm gives better results for most of the images.
wavelet transforms very few no of negative significant
4. Conclusions
Medical imaging generates vast amount of data . To
14
preserve these data, lossless compression is widely
preferred, because it allows the exact reconstruction of
COMPRESSION RATIO
12
10
the original image. In secured lossless image
8
compression, the proposed Region based Modified
6
SPIHT produces a lossless compressed embedded data
4
stream and the data stream is scrambled based on the
2
key to encrypt an image. The image scrambling scheme
0
51.58 60.44 53.38 54.14 52.14 58.91 38.13 36.14
has relatively simple implementation, low cost and low
PSNR in DB
delay operations which are highly desirable
characteristics for real time interactive applications. The
E Z W- A rit hC o de S P IH T - A rit hC scrambled data is further compressed by Modified
Runlength coding which is simple and a faster
S P IH T - M R LE P ro po s e d A p
technique. It not only compresses the image but also
increases the security level. Secured transmission helps
Figure 6. Comparison of various coding techniques in accurate diagnosis and research. The proposed
MRLE and Region based Modified SPIHT increases the
coefficients are found in first quadrant which decreases compression performance in secured lossless image
the PSNR ratio by a negligible factor between 0.0 DB compression. An excellent compression ratio is
and 0.0002 DB. achieved and also the quality of the image is found to be
good. MRLE is a fast, simple, lossless compression
Compression ratio of various images in table1 and its technique. It increases the compression performance
PSNR is shown in Fig 6. If we analyze the with an increase in security level.
decomposition results of (2,2) Daubechies transform
and S transform almost all the coefficients in first
Figure 3. Scrambled Lossless compression and Reconstruction of WILLIS Image result of (2,4) transform
CR=11.92
Figure 4. Scrambled Lossless compression and Reconstruction of Lena Image result of (2,4) transform CR=10.42
46
GVIP Journal, Volume 5, Issue 9, Dec. 2005
Hence the low cost faster and simpler implementation it is highly suitable for fast consultation and
highly satisfies the good candidate for telemedicine transmission of losslessly archived large medical
applications and also retains the option of secured images. I have implemented the algorithm for 2D
lossless transmission and archiving of the image. So images and I propose to extend this approach to 3D
images.
Figure 5. Scrambled Lossless compression and Reconstruction of Brain cora result of (2,4) transform CR=11.8
TABLE 1. Comparison on CR of different Integer wavelet transform of the proposed approach with other methods
Lena_gray 2.45 2.98 8.55 10.10 10.42 10.35 10.82 10.58 10.71 10.54 10.10
Blood cell 3.10 3.51 7.02 8.09 8.56 8.38 9.06 8.81 8.89 8.98 8.23
CT_skull 2.85 2.97 4.24 7.16 7.77 7.51 8.18 7.96 7.89 7.86 7.13
CT_wrist 4.80 4.79 5.88 8.41 8.83 9.08 9.33 9.49 9.21 9.30 8.52
Thorax 2.53 2.73 7.98 9.52 9.98 10.03 10.56 10.11 10.05 10.18 9.89
Spinal-gray scale 2.42 2.49 5.21 6.44 7.84 6.83 8.28 8.36 7.58 7.76 6.71
XRAY-grayscale 2.32 2.64 4.91 7.44 6.72 6.51 7.32 6.70 6.68 6.74 6.20
47
GVIP Journal, Volume 5, Issue 9, Dec. 2005
48