## Image Representation Based on Multi-Scale Edge Compensation (1999)

Venue: | in IEEE Internat. Conf. on Image Processing |

Citations: | 2 - 0 self |

### BibTeX

@INPROCEEDINGS{Xue99imagerepresentation,

author = {Xiaohui Xue and Xiaolin Wu},

title = {Image Representation Based on Multi-Scale Edge Compensation},

booktitle = {in IEEE Internat. Conf. on Image Processing},

year = {1999}

}

### OpenURL

### Abstract

In this paper we define an image model called MSEC (Multi-Scale Edge Compensation) and implement an image compression system based on MSEC. The idea is to represent the image by its multi-scale primal sketch and the background. The experimental results are positive; some important features of MSEC are proved. 1 Introduction 1.1 Problem Statement The mainstream of image compression research has been based on the Shannon Information Theory for years. JPEG, together with the various wavelet compression algorithms such as EZW [1] and SPIHT [2], consist of transform, quantization, and entropy coding. Transform is designed to represent and remove the statistical correlation within a certain spatial range in image data; quantization is to reduce the entropy of the transform coefficients; entropy coding is to approach the entropy of the quantized coefficients. All these algorithms are based mainly upon the Shannon Information Theory; they can be called statistical image coding. During the la...

### Citations

7146 |
A mathematical theory of communication
- Shannon
- 1948
(Show Context)
Citation Context ... can not be statistically compressed; just as the perpetual motion machine can never come true. Statistical redundancy has been revealed almost completely and accurately by Shannon information theory =-=[17]-=--[18]. Although the theory is elaborate, there are still a lot of difficulties in practice. Even in the simplest specific case of stationary source \Gamma it is well known that the conditional entropy... |

3436 | A Computational Approach To Edge Detection - Canny - 1986 |

1767 | Biorthogonal bases of compactly supported wavelets
- Cohen, A, et al.
- 1992
(Show Context)
Citation Context ...ven in one-dimension form for convenience, which does not lose the generality. 2.1 Definition MSEC scale transformation is borrowed from wavelet theory, for details of which the reader should consult =-=[30]-=-, [31], and [32]. In this section we use the notations as in [30]. Consider the biorthogonal wavelet system h = (h n ); g = (g n ); ~ h = ( ~ h n ); ~g = (~g n ). Denote the data sequence by c 0 = (c ... |

1321 |
Embedded image coding using zerotrees of wavelet coefficients
- Shapiro
- 1993
(Show Context)
Citation Context ...1 Problem Statement The mainstream of image compression research has been based on the Shannon Information Theory for years. JPEG, together with the various wavelet compression algorithms such as EZW =-=[1]-=- and SPIHT [2], consist of transform, quantization, and entropy coding. Transform is designed to represent and remove the statistical correlation within a certain spatial range in image data; quantiza... |

1156 |
Vision: A Computational Investigation into the Human Representation and Processing of Visual Information
- Marr
- 1982
(Show Context)
Citation Context ....2 Multi-Scale Zero Crossing Representation in Computer Vision That zero-crossing position of a Laplacian of Gaussian filtered image corresponds to the image edge is an important viewpoint of D. Marr =-=[9]-=-, which evolved into the multi-scale space decomposition method later [10]. Zero-crossing is of particular interest in computer vision and pattern recognition community. The main problem is on the com... |

751 |
Image coding using wavelet transform
- Antonini, Barlaud, et al.
- 1992
(Show Context)
Citation Context ...let from [31] (N=4); the filter h is 0.333, 0.807, 0.460, -0.135, -0.085, 0.035. ffl Biorthogonal wavelet from[30], pp557; the coefficients of the filters are as Tab. 1. ffl Biorthogonal wavelet B9-7 =-=[33]-=-. k 0 \Sigma1 \Sigma2 \Sigma3 \Sigma4 \Sigma5 \Sigma6 h 0.563 0.293-0.048-0.049 0.019 0.006-0.003 ~ h 0.560 0.296-0.047-0.055 0.022 0.011-0.006 Table 1: The filters coefficients of Example 3. Therefor... |

694 |
Arithmetic Coding for Data Compression
- Witten, Neal, et al.
- 1987
(Show Context)
Citation Context ...atically. For probability estimation method, we refer to [23]-[24]. Given the probability, arithmetic coding, instead of Huffman, is the most powerful entropy coding method in the sense of efficiency =-=[26]-=-. B.2 Discrete Stationary Memory Sources There are two choices for discrete stationary memory source. The one is to consider the joint entropy of the n-dimensional random vector H(X 1 ; X 2 ; \Delta \... |

561 |
Three approaches to the quantitative definition of information,” Probl
- Kolmogorov
- 1965
(Show Context)
Citation Context ...hannon Theory is far from the understanding of information in common sense. The former deals with the uncertainty in statistics. The latter needs to handle the vagueness of semantics and effect value =-=[19]-=-[20]. The statistical redundancy and the knowledge redundancy are closely related to the achievements and limitations of the Shannon Information Theory respectively. Thanks to the achievements of Shan... |

554 |
Characterization of signals from multiscale edges
- Mallat, Zhong
- 1992
(Show Context)
Citation Context ...more information, we refer to [13]. 1.3.3 Wavelet Maxima Algorithm Wavelet maxima algorithm reconstructs signals from the value and the position of the directional local maxim-as of wavelet transform =-=[14]-=--[16]. It was reported that for 256 \Theta 256 \Theta 8bits Lena, after 10 times iteration, the compression ratio is about 30 and PSNR is about 28dB. We consider wavelet maxima algorithm to be a succe... |

506 |
Multiresolution approximation and wavelet orthonormal bases of L2(R d
- Mallat
- 1989
(Show Context)
Citation Context ...sion form for convenience, which does not lose the generality. 2.1 Definition MSEC scale transformation is borrowed from wavelet theory, for details of which the reader should consult [30], [31], and =-=[32]-=-. In this section we use the notations as in [30]. Consider the biorthogonal wavelet system h = (h n ); g = (g n ); ~ h = ( ~ h n ); ~g = (~g n ). Denote the data sequence by c 0 = (c 0 n ). The decom... |

437 | Singularity detection and processing with wavelets - Mallat, HwangW |

390 |
Efficient Image Codec Based on Set Partitioning in Hierarchical Trees
- Said, Pearlman, et al.
- 1996
(Show Context)
Citation Context ...ement The mainstream of image compression research has been based on the Shannon Information Theory for years. JPEG, together with the various wavelet compression algorithms such as EZW [1] and SPIHT =-=[2]-=-, consist of transform, quantization, and entropy coding. Transform is designed to represent and remove the statistical correlation within a certain spatial range in image data; quantization is to red... |

200 |
Rate Distortion Theory
- Berger
- 1971
(Show Context)
Citation Context ...not be statistically compressed; just as the perpetual motion machine can never come true. Statistical redundancy has been revealed almost completely and accurately by Shannon information theory [17]-=-=[18]-=-. Although the theory is elaborate, there are still a lot of difficulties in practice. Even in the simplest specific case of stationary source \Gamma it is well known that the conditional entropy appr... |

155 |
Scale space filtering
- Witkin
- 1983
(Show Context)
Citation Context ...crossing position of a Laplacian of Gaussian filtered image corresponds to the image edge is an important viewpoint of D. Marr [9], which evolved into the multi-scale space decomposition method later =-=[10]-=-. Zero-crossing is of particular interest in computer vision and pattern recognition community. The main problem is on the completeness of multi-scale zero crossing representation and the stability of... |

112 |
Zero-crossings of a wavelet transform
- Mallat
- 1991
(Show Context)
Citation Context ...ension case, under certain conditions, the signal can be reconstructed by the multi1scale zero crossing representation [11], and the reconstruction can be stabilized by some supplementary information =-=[12]-=-. For more information, we refer to [13]. 1.3.3 Wavelet Maxima Algorithm Wavelet maxima algorithm reconstructs signals from the value and the position of the directional local maxim-as of wavelet tran... |

87 |
Second generation image coding techniques
- Kunt, Ikonomopoulos, et al.
- 1985
(Show Context)
Citation Context ...ded multi-scale primal sketch and the background. 1.3 Brief Review of Related Work 1.3.1 Second Generation Image Compression In 1980s, M. Kunt put forward the second-generation image coding technique =-=[3]-=--[7]. Two classes of methods are provided; one is based upon image segmentation, the other is upon directional decomposition. We consider the idea expressed by the second-generation image compression ... |

70 |
On the modeling of DCT and subband image data for compression
- Birney, Fischer
- 1995
(Show Context)
Citation Context ...h \Gammalog 2 p(i). There are still some important problems left, e.g. the probability can be calculated locally or globally, dynamically or statically. For probability estimation method, we refer to =-=[23]-=--[24]. Given the probability, arithmetic coding, instead of Huffman, is the most powerful entropy coding method in the sense of efficiency [26]. B.2 Discrete Stationary Memory Sources There are two ch... |

41 | Recent results in high-compression image coding - Kunt, Bernardi, et al. - 1987 |

38 |
Reconstruction from zero crossings in scale space
- Hummel, Moniot
- 1989
(Show Context)
Citation Context ...he signal can be reconstructed by the multi1scale zero crossing representation [11], and the reconstruction can be stabilized by some supplementary information [12]. For more information, we refer to =-=[13]-=-. 1.3.3 Wavelet Maxima Algorithm Wavelet maxima algorithm reconstructs signals from the value and the position of the directional local maxim-as of wavelet transform [14]-[16]. It was reported that fo... |

28 |
B.F.: Information in the zero crossings of bandpass signals
- Logan
- 1977
(Show Context)
Citation Context ...e zero crossing representation and the stability of the reconstruction. In onedimension case, under certain conditions, the signal can be reconstructed by the multi1scale zero crossing representation =-=[11]-=-, and the reconstruction can be stabilized by some supplementary information [12]. For more information, we refer to [13]. 1.3.3 Wavelet Maxima Algorithm Wavelet maxima algorithm reconstructs signals ... |

11 |
Predictive coding
- Elias
- 1955
(Show Context)
Citation Context ... model ( S(p(a 1 ); p(a 2 ); \Delta \Delta \Delta ; p(a m )) p(a j ) = P L i=1 P i p(a j ji) (54) It was proved that the entropy of the composition model is smaller than that of the average model [27]=-=[28]-=-. So, for a non-stationary source, we have 1. The key to improve the coding efficiency is the recognition of stationary sub-source. C Formalization of MSEC Model This section formalizes MSEC model for... |

9 | High compression image coding via directional filtering - Ikonomopoulos, Kunt - 1985 |

8 |
Edge representation from wavelet transform maxima
- Zhong
- 1991
(Show Context)
Citation Context ...information, we refer to [13]. 1.3.3 Wavelet Maxima Algorithm Wavelet maxima algorithm reconstructs signals from the value and the position of the directional local maxim-as of wavelet transform [14]-=-=[16]-=-. It was reported that for 256 \Theta 256 \Theta 8bits Lena, after 10 times iteration, the compression ratio is about 30 and PSNR is about 28dB. We consider wavelet maxima algorithm to be a successful... |

7 |
On the Performance of a Contour Coding Algorithm
- Eden, Kocher
- 1985
(Show Context)
Citation Context ...multi-scale primal sketch and the background. 1.3 Brief Review of Related Work 1.3.1 Second Generation Image Compression In 1980s, M. Kunt put forward the second-generation image coding technique [3]-=-=[7]-=-. Two classes of methods are provided; one is based upon image segmentation, the other is upon directional decomposition. We consider the idea expressed by the second-generation image compression algo... |

6 |
Synthetic highs, an experimental tv bandwidth reduction system
- Schreiber, Knapp, et al.
- 1959
(Show Context)
Citation Context ...ional decomposition. We consider the idea expressed by the second-generation image compression algorithm is in advance for its time, and it is still worthy today. Let us mention synthetic high method =-=[8]-=- here, the idea of which is much more simple but equally significant. 1.3.2 Multi-Scale Zero Crossing Representation in Computer Vision That zero-crossing position of a Laplacian of Gaussian filtered ... |

5 |
Probability estimation in arithmetic and adaptive huffman entropy coders
- Duttweiler, Chamzas
- 1995
(Show Context)
Citation Context ...mmalog 2 p(i). There are still some important problems left, e.g. the probability can be calculated locally or globally, dynamically or statically. For probability estimation method, we refer to [23]-=-=[24]-=-. Given the probability, arithmetic coding, instead of Huffman, is the most powerful entropy coding method in the sense of efficiency [26]. B.2 Discrete Stationary Memory Sources There are two choices... |

4 |
Digital Image Compression Based on Visual Perception
- Glenn
- 1993
(Show Context)
Citation Context ...psychovisual redundancy is considerably large. W. Glenn estimated that at least 50 times compression is available by taking advantage of the psychovisual factors in human visual system in proper ways =-=[21]-=-. It was also theoretically proved that the lower bound of the rate-distortion function would be decreased by one or two orders of magnitude [22]. In a certain sense, lossless compression cannot make ... |

2 | Sketch Based Coding of Gray Level Images - Carlsson - 1988 |

2 |
The Application of Human Visual System Model to Digital Color Images Compression
- Hall
- 1983
(Show Context)
Citation Context ...hovisual factors in human visual system in proper ways [21]. It was also theoretically proved that the lower bound of the rate-distortion function would be decreased by one or two orders of magnitude =-=[22]-=-. In a certain sense, lossless compression cannot make use of psychovisual redundancy; while lossy compression must do. Psychovisual factors have been partly taken into account in many compression met... |

1 |
Information Science and Information Theory
- Zhong
- 1990
(Show Context)
Citation Context ...on Theory is far from the understanding of information in common sense. The former deals with the uncertainty in statistics. The latter needs to handle the vagueness of semantics and effect value [19]=-=[20]-=-. The statistical redundancy and the knowledge redundancy are closely related to the achievements and limitations of the Shannon Information Theory respectively. Thanks to the achievements of Shannon ... |

1 | Bayesian Approach to a Family of Fast Attack Priors for Binary Adaptive Codings - Zandi, Langdon - 1992 |

1 |
Advances in the Coding of Television Signals
- Yao
- 1986
(Show Context)
Citation Context ...ately, namely H(S) +H(T ) = H(S; T ) = H(X; Y )sH(X)+ H(Y ) (52) B.4 Non-stationary Source Non-stationary source S can be modelled as the composition of a series of stationary subsourcessfS i g L i=1 =-=[27]-=-. Within a particular interval, the output of the composition model is merely that of a particular S i , which produces a segment of stationary signal. Suppose the alphabets of S are fa 1 ; a 2 ; \Del... |

1 |
Structural Pattern Recognition, Springer-Verlag
- Pavilidis
- 1977
(Show Context)
Citation Context ...it is more appropriate to represent them with numerical approximation methods, which is exactly the problem of the representation of two-dimension structure in computer vision and pattern recognition =-=[29]-=-. The accuracy of representation there must meet the demand of the subsequent processing or recognition. Now we need more precise approximation to satisfy human visual system (HVS). However, in order ... |