New region of interest image coding using general layered bitplane shift for medical image compression.
Keywords: medical image coding, JPEG2000, region of interest, bitplane shift
Medical images are useful tools for diagnostic investigation. For example, the CT or MRI image, which produce human body pictures in digital form. However, their usage may be made difficult because of the amount of data to store or because of the duration of communication over a limited capacity channel. Lossy medical image compression techniques may have the potential for widespread medical acceptance, but the compression will reduce the image fidelity, especially when the images are compressed at lower bit rates. So a new image compression scheme called region of interest (ROI) coding was presented in -.
The basic idea of ROI coding is that certain parts of the image that are of higher importance than others are encoded at higher quality than the background (BG) and are transmitted first or at a higher priority -. Recently, two basic coding strategies are presented in the literatures-ROI coding based on zerotree or zeroblock scheme and ROI coding based on bitplane coding and EBCOT-. The former have been researched and applied widely, as for example EZW, SPIHT and SPECK for volumetric medical data -. However, the methods have high coding complexity and do not realize spatial scalability. The latter can realize spatial scalability and reduces coding complexity, which enable it to be recommended by JPEG2000.
JPEG2000 is the new state-of-the-art image compression standard designed for broad range of applications, including the compression and transmission of medical images. The new standard is based on wavelet technology and a layered file format that offers lossless compression, diagnostic-quality lossy compression and advanced system level functionality. JPEG2000 has been selected for inclusion in the DICOM standard for medical image transfer (DICOM Working Group 4-Image Compression Group).
In JPEG2000 standard, two coding algorithms so-called Maxshift (maximum shift) method in part 1 and the general scaling-based method in part 2 along with the syntax of a compressed codestream are presented. The Maxshift method exploits the possibility of representing an arbitrarily shaped ROI by shifting the ROI coefficient values on the top of the BG coefficient range. This actually produces a separate ROI-BG bit-stream. The scaling based method realizes a coefficient ranking by using a subband coefficient weighting mask prior to the quantization stage, producing a bit stream where ROI and BG are coded together with a different quality level. In these methods, regions of interest can have better quality than the rest at any decoding bit-rate , .
In this paper, we presented a new and efficient ROI coding method for medical image based on bitplane shift scheme, which is called general layered bitplane shift (GLBShift) method. The proposed method can delivers much more flexibility than Maxshift method in the adjustment of compression quality in ROI and BG. The experiment results for CT and MRI images show that GLBShift method has four primary advantages: (1) the proposed method has the flexibility for an arbitrary scaling value to define the relative importance of the ROI and the BG wavelet coefficients; (2) it can support some BG bitplanes are prior to encode if the ROI detail is not important; (3) it can support arbitrarily shaped multiple ROI coding with different degrees of interest without coding ROI shape information; (4) it can decrease the risk of bit-streams overflow.
The remainder of this paper is structured as follows. In Section 2, the several main ROI coding methods for the medical image are reviewed. In Section 3, the GLBShift method is presented. In Section 4, the multiple ROIs coding based on the GLBShift method is proposed. In Section 5, we present the complexity comparison among Maxshift, BbBShift and the proposed GLBShift method, while the experimental results for medical images are given in Section 6. Finally, the conclusions are drawn in Section 7.
II. Several ROI coding methods for the medical image
ROI image coding is a new feature in JPGE2000, which allows ROI to be coded with better quality than the rest of an image. Two kinds of ROI coding methods are included in the standard: the Maxshift method and the general scaling based method.
A. The general scaling based method
As illustrated in Fig. 1(a), the general scaling based method can place ROI associated bits in the higher bitplanes by downshifting the bits of BG coefficients from Most Significant Bitplane (MSB) to Least Significant Bitplane (LSB), so that ROI coefficients can be coded firstly after the wavelet transform. As any scaling value is supported, the general scaling based method allows fine control on the relative importance between ROI and BG. However, the general scaling based method has three major drawbacks -.
[FIGURE 1 OMITTED]
* First, it needs to encode and transmit the shape information of the ROIs. This rapidly increases the encoding and decoding complexity.
* Second, if arbitrary ROI sharps are desired, the shape coding will consume a large number of bits, which significantly decreases the overall coding efficiency.
* Third, it is not convenient to deal with different wavelet subbands in different ways, which is sometimes desired by the users.
So the current standard restricts ROI shape to be rectangle and ellipse .
B. Maxshift method
Maxshift method proposed by JPEG2000 can solve these problems in the general scaling based method effectively. In other words, it is a particular case of the general scaling-based method when the scaling value is so large that there is no overlapping between BG and ROI bitplanes. So the scaling value, s, must satisfy (1):
s [greater than or equal to] max([M.sub.b]) (1)
Where max ([M.sub.b]) is the largest number of magnitude bitplanes for any coefficient. In that case, all significant bits associated with the ROI coefficients after scaling will be in higher bitplanes than all the significant bits associated with the BG coefficients. Therefore, ROI shape is implicit for the decoder in this method, and arbitrarily shaped ROI coding can be supported. Fig. 1(b) shows these two ROI coding methods in JPEG2000.
Five major limitations of Maxshift method for ROI coding are as follows -.
* First, it does not have the flexibility for an arbitrary scaling value to define the relative importance of the ROI and the BG wavelet coefficients. This means in all the subbands, no information about the non-ROI coefficients can be received until every detail of the ROI coefficients has been fully decoded, even if the detail is imperceptible random noise or unnecessary information.
* Second, this method requires decoding of all ROI coefficients before accessing bitplanes of the background and uses large shifting values that significantly increase the number of total bi-planes to encode.
* Third, when there are multiple ROIs in the same image, any ROI cannot have its own scaling value. Therefore, different priority during encoding and transmission of the image.
* Fourth, as all ROI bitplanes are higher than all BG bitplanes, the scaling value rapidly increase the risk of bit-stream overflow.
* Fifth, it is inflexible in interactive net browser and not full to use processing transmitting of JPEG 2000.
C. ROI coding based on RBDWT and OB-SPIHT
Region of interest (ROI) coding techniques are particularly suitable for medical imaging. Such methods provide the possibility of adequately compressing those regions with diagnostic relevance with better quality than the rest of the image. In  and , W. A. Pearlman proposed OB-SPIHT coding algorithm. It is object-based extensions of full-image SPIHT algorithms. It preserves the features of the original methods. When OB-SPIHT is used with RBDWT, it is possible to efficiently perform wavelet subband decomposition of a region of arbitrary shape, while maintaining the same number of wavelet coefficients to be coded than pixels within the region and keeping spatial correlation and self-similarity across subbands. The main drawback of RBDWT in the ROI coding is that the shape information of ROI must be coding, which significantly decreases the ROI coding efficiency.
III. General layered bitplane shift method for medical image compression
A. Improved methods for JPEG2000 ROI image coding
Because of the limitations of two standard ROI coding algorithms, some improved methods for ROI coding were proposed. A new method was proposed in  with low scaling values to take advantages of two standard methods. It is implemented by removing all the overlapping bitplanes between ROI and BG coefficients, which relatively modified the quantization steps of coefficients. However, the method brought the reduction of final ROI and BG qualities.
A bitplane-by-bitplane shift (BbBShift) method was proposed in  by shifting the bitplanes on a bitplane-by-bitplane basis instead of shifting them all at once in Maxshift method. Although it supports arbitrarily shaped ROI coding without coding shapes, it is difficult for the BbBShift method to code multiple ROIs with different priority during encoding and transmission. Fig. 2 shows the scaling scheme based on BbBShift method.
[FIGURE 2 OMITTED]
The partial significant bitplanes shift (PSBShift) method proposed by  shifts part of the most significant of ROI coefficients instead of shifting the whole bitplanes as the standard methods do. But this method cannot decode all ROIs coefficients before all BG coefficients are decoded because some residual bitplanes cannot be shifted at the encoder. An Up-Down Bitplanes Shift scheme is presented in .
In this paper, a flexible bitplanes shift coding method using general layered bitplane shift (GLBShift) is proposed, which can efficiently solve the five problems in Maxshift method and ensure all ROIs to be decoded before BG is decoded for medical images.
B. GLBShift scheme
The ROI coding system based on the GLBShift method includes three parts: integer wavelet transform (IWT), ROI mask generation and the bitplane scaling encoding based on GLBShift method.
1) Integer wavelet transform
Because the wavelet image coefficients by the general discrete wavelet transforms (DWTs) are floating-point numbers, the efficient lossless coding is not possible using DWT. In this case, Sweldens and Calderbank presented the reversible integer-to-integer wavelet transforms based on the Lifting Scheme (LS) . In LS, the IWT can be described through polyphase matrix using Euclidean Algorithm as
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)
P(z) can be defined as analysis filters. [S.sub.i](Z) and [t.sub.i](Z) can be defined as Laurent polynomials. K is the lifting factor and is a constant.
Since using IWT can reconstruct a perfectly lossless decoding image, the image coding based on IWT is very significant for medical and remote sense image compression. Additionally, the use of IWT is also a means to reduce the memory demands of the compression algorithm as integers are used instead of real numbers. IWT can be composed of split, forecast and update by lifting scheme. The details of IWT are introduced in  and .
2) ROI mask generation
When an image is coded with one or multiple ROIs, it should be possible to reconstruct the entire ROI at a higher bit rates than BG part. It is therefore necessary to identify the wavelet coefficients needed for the reconstruction of the ROIs, so that they can be coded at a higher quality. An ROI mask will be created for this purpose. This mask indicates which wavelet coefficients belong to an ROI and to which one if multiple ROIs are defined.
In fact, if all coefficients in the ROI mask are losslessly encoded, the ROI can be losslessly reconstructed. This is performed for lines and columns at each decomposition level. The process is then repeated for the remaining levels, until the entire wavelet tree is processed.
For multiple ROIs, one additional issue arises. Since the ROI mask expends as it is created, there is a possibility for two ROIs that are disjoint in the spatial domain to have some have some coefficients in common in the wavelet domain. The most common solutions are to redundantly encode such coefficients or to them coefficients in the highest quality ROI only , . Fig. 3 shows the ROI mask of a MRI image. We select three ROIs and two level wavelet decomposition. Fig. 4 gives the ROI mask of a CT image with a ROI. The number of wavelet decomposition is three.
[FIGURES 3-4 OMITTED]
3) The scaling strategies of GLBShift
In the new ROI coding system, the wavelet transform is firstly performed, and the transformed coefficients are eventually quantized. Then, the certain number bitplanes of ROI or BG are up-shifted. Finally, the arithmetic coding for all bitplanes of ROI and BG are completed.
The new method divides all bitplanes into three parts: most significant bitplanes of ROI and BG coefficients (MB), general significant bitplanes of ROI and BG coefficients (GB), and least significant bitplanes of ROI and BG coefficients (LB).
As illustrated in Fig. 4, instead of shifting the bitplanes all at once by same scaling value s as in Maxshift, three new scaling strategies are preformed in GLBShift method as follow:
* First, these ROI bitplanes belonged to LB are up-shifted until no overlapping happens between BG and ROI bitplanes.
* Second, these ROI and BG bitplanes belonged to GB are up-shifted over the top of the maximum ROI bitplane in LB. However, in GB, no bitplane is shifted between ROI and BG.
* Third, the most significant ROI and BG bitplanes are up-shifted to the higher position than the maximum bitplane in GB. They are defined as MB region. These ROI bitplanes in MB must be no overlapping with all BG bitplanes belonged to MB.
Based on these scaling strategies, an illustration of GLBShift method is shown in Fig. 4.
C. The scaling steps of GLBShift
In this paper, we index the original bottom bitplane (original LSB) as bitplane 1, the next to original bottom as bitplane 2, and so on. At the encoder, the basic parameters of GLBShift method are defined as following:
* [S.sub.1]: The bitplane number in MB region belonged to ROI coefficients;
* [S.sub.2]: The bitplane number in MB region belonged to BG coefficients;
* [S.sub.3]: The bitplane number in GB region belonged to ROI or BG coefficients;
* [S.sub.4]: The bitplane number in LB region belonged to ROI coefficients;
* [S.sub.5]: The bitplane number in LB region belonged to ROI coefficients;
According to Fig. 4 and the definition of parameters, we conclude that [S.sub.1], [S.sub.2], [S.sub.4] and [S.sub.5] must satisfy (3) and (4):
[S.sub.1] = [S.sub.2] (3)
[S.sub.4] = [S.sub.5] (4)
At the encoder, the bitplane shift algorithm can be described as follows:
1. For any bitplane b belonged to an ROI coefficient:
1.1. If original bitplane b satisfies that b belongs to LB, shift it up to bitplane b + [S.sub.5];
1.2. If original bitplane b satisfies that b belongs to GB,, shift it up to bitplane b + [S.sub.5];
1.3. If original bitplane b satisfies that b belongs to MB,, shift it up to bitplane b + [S.sub.2] + [S.sub.5];
2. For any bitplane b belonged to an BG coefficient:
2.1. If original bitplane b satisfies that b belongs to LB, no shift and encoding b directly;
2.2. If original bitplane b satisfies that b belongs to GB,, shift it up to bitplane b + [S.sub.5];
2.3. If original bitplane b satisfies that b belongs to MB, shift it up to bitplane b + [S.sub.5].
At the decoder, for any given non-zero wavelet coefficient, the first step is to identify whether it is a bitplane of the ROI coefficient or the BG coefficient. The can be done by examining the bitplane level of its MSB. The decoding algorithm is presented as follows:
1. If b > [S.sub.2] + [S.sub.3] + [S.sub.4] + [S.sub.5], then b [member of] ROI and b [member of] MB, shift it down to bitplane b - [S.sub.2] - [S.sub.5];
2. If [S.sub.3] + [S.sub.4] + [S.sub.5] < b [less than or equal to] [S.sub.2] + [S.sub.3] + [S.sub.4] + [S.sub.5], then b [member of] MB and b [member of] BG, shift it down to bitplane b - [S.sub.5];
3. If [S.sub.4] + [S.sub.5] < b [less than or equal to] [S.sub.3] + [S.sub.4] + [S.sub.5], then b [member of] GB, but b [member of] ROI or b [member of] BG, shift it down to bitplane b - [S.sub.5];
4. If [S.sub.5] <b [less than or equal to] [S.sub.4] + [S.sub.5], then b [member of] LB and ROI, shift it down to bitplane b - [S.sub.5];
5. If b [less than or equal to] [S.sub.5], then b [member of] LB and b [member of] BG, no shift and decoding b directly.
IV. Multiple ROI coding using GLBShift method
In JPEG2000, both Maxshift method and the general scaling based method can support the multiple ROI coding. However, each method has itself the drawbacks. There are two major drawbacks of Maxshift method for multiple ROI coding , .
* First, the coefficient bitplanes of all ROIs must be scaled with the same values, which does not have the flexibility to allow for an arbitrary scaling value to define the relative importance of the ROIs and BG wavelet coefficients.
* Second, it does not have the flexibility to allow for an arbitrary scaling value to define the degree of relative importance between the ROI and the BG wavelet coefficients.
The general scaling based method can offer the multiple ROIs coding with different degrees of interest, but it has three major drawbacks. Firstly, it needs to encode the shape information of ROIs. This shape information significantly increases the complexity of encoder/decoder when the number of the ROIs increases. Secondly, when arbitrary ROI shapes are desired, the shape coding of the ROIs will consume a large number of bits, which reduces the overall coding efficiency. The current standard in JPEG2000 attempts to avoid this problem and only defines rectangle or ellipse shaped ROIs because they can be coded with a small number of bits. Finally, it is not convenient to deal with different wavelet subbands according to different degrees of interest, which is sometimes is very important to code and transmit for objectors ,.
To solve these problems in Maxshift and the general scaling based method, we propose GLBShift method for multiple ROI coding. The scheme of the GLBShift method for multiple ROI coding is illustrated in Fig. 5. As illustrated in Fig. 5, the encoding and decoding methods for multiple ROI are similar to those for single ROI. However, a key point must be noticed. For all ROIs, the shifted ROI bitplane numbers in MB can be different. According to the requirement of regions of relative importance among ROIs, we can adjust the shifted ROI bitplane numbers in MB. There is only a scaling value ([S.sub.1=MAX]) for all ROIs in MB. Based on these above facts, the GLBShift method not only can support arbitrary ROIs shape without shape coding, but also allows arbitrary scaling value between the ROIs and BG, which enables the flexible adjustment of compression quality in ROIs and BG according to different degrees of interest. Additionally, the proposed method can decrease the bitplane scaling value of ROIs, which can decrease the risk of bit-stream overflow in multiple ROI coding.
[FIGURE 5 OMITTED]
At low bit rates, different bitplanes are decoded with different degrees of relative interest. At high bit rates, both ROIs and BG can be coded with high quality and the difference among them is not very noticeable. Additionally, the new method can also support some BG bitplanes are prior to encode if the detail of ROIs is imperceptible random noise or not important.
V. Complexity analysis
Since it is not necessary for the GLBShift method to code and transmit the shape information of every ROI, the complexity of the proposed algorithm is less and coding efficiency is higher than the general scaling based method for the same image.
Compared with Maxshift method, a more complicated procedure is included in GLBShift method so that all bitplanes are shifted to the correct positions. In the decoder, the more complex scaling procedure is needed to reconstruct the original bitplanes. The codestream generated by the GLBShift method is not compliant with the current JPEG2000 format.
Compared with the PSBShift and BbBShift method, the presented method is simpler than BbBShift in both encoder and decoder. And it is slightly more complex than PSBShift. However, GLBShift can ensure that all bitplanes of the ROI coefficients are decoded before all bitplanes of the BG coefficients are decoded.
For multiple ROI coding, GLBShift cannot only support different degrees of interest at different decoding rates, but also dose not limit the ROIs shapes. So the proposed method has lower complexity than the general scaling based method and BbBShift method. The complexity is similar to the PSBShift method in . Therefore, PBLShift method has more advantages for multiple ROI coding than other ROI methods.
VI. Experimental results
We select a lot of CT and MRI images for the ROI coding and compare the compression results of GLBShift method with those of other efficient ROI coding methods. In all experiments, we utilize the 5/3 integer wavelet filter group, which is one of the most efficient and is recommended by JPEG2000 standard. The number of decomposed levels is five.
Fig. 6 gives a 512 x 512 gray CT image and two reconstructed CT images with single ROI. The original medical image has a rectangular ROI, whose covering area is about 10.3% of the whole image. We present the ROI mask with five-level wavelet decomposition. The reconstructed CT image using Maxshift method and GLBShift method are show and the decoding rate is 1.0bpp.
[FIGURE 6 OMITTED]
Fig. 7 gives an original MRI image, a decomposed MRI image with an arbitrarily shaped ROI and two reconstructed MRI images. The image is a 512 x 512 gray MRI image and the selected area is about 6.9% of the whole image. We compare the coding performance of GLBShift method with that of Maxshift method at 1.0 bpp.
[FIGURE 7 OMITTED]
As illustrated Fig. 6 and Fig. 7, it can be observed that without visual difference at the ROI between the reconstructed medical image using Maxshift and that using GLBShift. However, GLBShift coded images provides better quality at the BG, especially at low bit rates such as 0.5bpp and 1.0bpp.
Figure 8 shows original and reconstructed MRI images with three ROIs using different ROI coding methods. The MRI is a 512 x 512 gray image. The covering area of ROI-1 is about 2.46% of the whole image, that of ROI-2 is 2.26% and that of ROI-3 is 3.20%. Fig. 8 (b) shows the reconstructed MRI image using Maxshift at 1.0bpp. Fig. 8 (c) shows the reconstructed MRI image using GLBShift at 0.5bpp and Fig. 8 (d) is the reconstructed MRI image using GLBShift at 1.0bpp. The ROI-1 is the left and top part, the ROI-2 is the bottom part and the ROI-3 is the right and top part.
[FIGURE 8 OMITTED]
Experimental results for medical image show that the presented method can efficiently code multiple ROIs based on different degrees of interest without any shape information of the ROIs.
ROI coding for medical applications is a very desirable feature, such as telemedicine, volumetric medical data compression and medical image analysis. The proposed algorithm is highly flexible. It allows the user to request an ROI at any moment, to switch between different ROIs, and to switch between ROI and BG transmission. It also supports for lossless coding of ROIs only or the entire image. These features leave room for network server optimization.
GLBShift method has four major advantages for medical image coding. First, the new method has the flexibility for an arbitrary scaling value to define the relative importance of the ROI and the BG wavelet coefficients. This means the better quality of BG can be provided by GLBShift than Maxshift. Second, GLBShift can support some BG bitplanes are prior to encode if the ROI detail is not important. Third, it can support arbitrarily shaped multiple ROI coding with different degrees of interest without coding the ROI shapes, which is very important to interactive network medical image transmission and the distance diagnosis based on large images. Fourth, it can decrease the risk of bit-streams overflow. We expect this idea is valuable for future medical image coding.
The authors would like to express their gratitude to the anonymous reviewers for their useful comments and thank all of the participants in the subjective testing for their time and effort.
This paper is supported in part by Project Research on Arbitrary Shaped Multiple Regions of Interest Image Coding Methods based on different degrees of interest supported by National Natural Science Foundation of China (No. 60602035) and the Youth Teacher Foundation of Beijng Normal University.
 ISO/IEC 15444-1, "Information technology-JPEG2000 image coding system-Part 1: Core coding system," Final Publication Draft, 2002.
 C. Christopoulos, J. Askelf and M. Larsson, "Efficient methods for encoding regions of interest in the upcoming JPEG 2000 still image coding standard", IEEE Signal Processing Letters, 7(9), pp. 247-249, 2000.
 A. Skodras, C. A. Christopoulos and T. Ebrahimi, "The JPEG 2000 still image compression standard", IEEE Signal Processing Magazine, 12(9), pp.36-58, 2001.
 ISO/IEC 15444-2, "Information technology-JPEG2000 image coding system--Part 2: Extensions," Final Publication Draft, 2002.
 C. A. Christopoulos, A. Skodras and T. Ebrahimi, "The JPEG2000 still image coding system: An overview", IEEE Trans. Consumer Electronics, 46(4), pp. 1103-1127, 2000.
 M. Penedo, W. A. Pearlman, P. G. Tahoces, M. Souto, and J. J. Vidal, "Region-Based Wavelet Coding Methods for Digital Mammography", IEEE Trans. Medical Imaging, 22(10), pp. 1288-1296. 2003.
 I. Ueno, W.A. Pearlman, "Region of Interest Coding in Volumetric Images with Shape-Adaptive Wavelet Transform", In Proceedings of the SPIE/IS&T Electronic Imaging 2003, 5022, pp. 1048-1055, 2003.
 A. Said and W.A. Pearlman, "New Fast and Efficient Image Codec Based on Set Partitioning in Hierarchical Trees", IEEE Trans, Circuit System Video Technology, 6(3), pp.243-249, 1996.
 W. A. Pearlman, A. Islam, N. Nagaraj, A. Said, "Efficient, low-complexity image coding with a set-partitioning embedded block coder", IEEE Trans. Circuit and Systems for Video Technology, 14(11), 1219-1235, 2004.
 L. Liu and G. Fan, "A new JPEG 2000 Region of interest image coding method: Partial Significant bitplanes shift", IEEE Signal Processing Letters, 10(2), pp.35-38, 2003.
 Z. Wang and A. C. Bovik, "Bitplane-by-Bitplane shift (BbBShift)-a suggestion for JPEG 2000 region of interest image coding", IEEE Signal Processing Letters, 9(5), pp.160-162, 2002.
 R. Grosbois, D. S. Cruz and T. Ebrahimi, "New approach to JPEG 2000 compliant region-of-interest coding", In Proceedings of the SPIE 46th Annual Meeting, Applications of Digital Image Processing, San Diego, CA, pp.267-275, 2001.
 L-b. Zhang and K. Wang. "Regions of Interest Image Coding Using Up-Down Bitplanes Shift For Network Applications", Lecture Notes in Computer Science, 3222, pp. 513-516, 2004.
 I. Daubechies and W. Sweldens, "Factoring wavelet transforms into lifting steps", Journal of Fourier Analysis Application, 4(3), pp.247-269, 1998.
 R.C. Calderbank, I. Daubechies and W. Sweldens, "Lossless image compression using integer to integer wavelet transforms", In Proceeding of IEEE International Conference on Image Processing, pp. 596-599, 1997.
 L-b. Zhang. "A New Region of Interest Image Coding for Narrowband Network: Partial Bitplane Alternating Shift", Lecture Notes in Computer Science, 3779, pp. 425-432, 2005.
 L-b. Zhang and K. Wang. "New Region of Interest Image Coding and its Applications for Remote Sensing Image", Chinese Optics Letter, 4(2), pp. 76-79, 2006.
Li-bao Zhang was born in Hebei, China, in 1977. He received the B.S. degree in College of Information Science and Engineering from Jilin University of Technology, Changchun, Jilin, China, in 1999. He received the M.S. and Ph.D. degrees in College of Communication Engineering from Jilin University, Changchun, Jilin, China, in 2002 and 2005, respectively.
Since August 2005, he has taught in College of Information Science and Technology from Beijing Normal University, Beijing, China. His research interests include digital signal processing, wavelets, image coding and multimedia systems.
Xian-chuan Yu was born in Sichuan, China, in 1968. He received the B.S., M.S. and Ph.D. degrees from Jilin University, Changchun, Jilin, China, in 1989, 1992, and 1995, respectively.
Since June 1998, he has taught in Beijing Normal University, Beijing, China. Now, he is teaching as professor in College of Information Science and Technology from Beijing Normal University. His research interests include digital signal processing, remote sensing image processing and data mining.
Li-bao Zhang (1) and Xian-chuan Yu (2)
(1) College of Information Science and Technology, Beijing Normal University, Xinjiekouuwai Street 19, Beijing, Chin email@example.com
(2) College of Information Science and Technology, Beijing Normal University, Xinjiekouwai Street 19, Beijing, China firstname.lastname@example.org
|Printer friendly Cite/link Email Feedback|
|Author:||Zhang, Li-bao; Yu, Xian-chuan|
|Publication:||International Journal of Computational Intelligence Research|
|Date:||Jan 1, 2007|
|Previous Article:||Application of neural networks to business bankruptcy analysis in Thailand.|
|Next Article:||Intrusion detection by backpropagation neural networks with sample-query and attribute-query.|