## Image information and visual quality (2004)

Venue: | IEEE Trans. Image Processing |

Citations: | 140 - 30 self |

### BibTeX

@INPROCEEDINGS{Sheikh04imageinformation,

author = {Hamid R. Sheikh and Alan C. Bovik},

title = {Image information and visual quality},

booktitle = {IEEE Trans. Image Processing},

year = {2004},

pages = {430--444}

}

### Years of Citing Articles

### OpenURL

### Abstract

Measurement of image quality is crucial for many imageprocessing algorithms. Traditionally, image quality assessment algorithms predict visual quality by comparing a distorted image against a reference image, typically by modeling the Human Visual System (HVS), or by using arbitrary signal fidelity criteria. In this paper we adopt a new paradigm for image quality assessment. We propose an information fidelity criterion that quantifies the Shannon information that is shared between the reference and the distorted images relative to the information contained in the reference image itself. We use Natural Scene Statistics (NSS) modeling in concert with an image degradation model and an HVS model. We demonstrate the performance of our algorithm by testing it on a data set of 779 images, and show that our method is competitive with state of the art quality assessment methods, and outperforms them in our simulations. 1.

### Citations

9359 |
Elements of information theory
- Cover, Thomas
- 1991
(Show Context)
Citation Context ...note as . With the stated assumptions on and the distortion model (2), we get (6) (7) 436 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 2, FEBRUARY 2006 (8) (9) where we get (6) from chain rule =-=[35]-=- and (7) from the conditional independence of and given , and denotes the determinant. Similarly, we can show that for the test image (10) (11) Since is symmetric, it can be factored as , where is an ... |

1334 |
Embedded image coding using zerotrees of wavelet coefficients
- Shapiro
- 1993
(Show Context)
Citation Context ...e (and feasible) understanding is still lacking. NSS modeling may serve to fill this gap. NSS have been explicitly incorporated into a number of image processing algorithms: in compression algorithms =-=[21]-=-–[24], denoising algorithms [25]–[27], image modeling [28], image segmentation [29], and texture analysis and synthesis [30]. While the characteristics of the distortion processes have been incorporat... |

398 | efficient image codec based on set partitioning in Hierarchical trees - Said, Pearlman, et al. - 1996 |

391 | Image denoising using scale mixtures of gaussians in the wavelet domain
- Portilla, Strela, et al.
- 2003
(Show Context)
Citation Context ...ng an appropriate distribution for , and that the nonlinear dependencies between the wavelet coefficients of natural images can be captured by modeling the field as being highly self-correlated [27], =-=[34]-=-. B. Distortion Model The purpose of a distortion model is to describe how the statistics of an image are disturbed by a generic distortion operator. The distortion model that we have chosen provides ... |

386 |
The visible differences predictor: an algorithm for the assessment of image fidelity
- Daly
- 1993
(Show Context)
Citation Context ... and computational feasibility. A detailed discussion of these methods can be found in [4]–[6]. A number of HVS-based methods have been proposed in the literature. Some representative methods include =-=[7]-=-–[14]. B. Arbitrary Signal Fidelity Criteria Researchers have also attempted to use arbitrary signal fidelity criteria in a hope that they would correlate well with perceptual quality. In [15] and [16... |

315 | A parametric texture model based on joint statistics of complex wavelet coefficients
- Portilla, Simoncelli
- 2000
(Show Context)
Citation Context ...ted into a number of image processing algorithms: in compression algorithms [21]–[24], denoising algorithms [25]–[27], image modeling [28], image segmentation [29], and texture analysis and synthesis =-=[30]-=-. While the characteristics of the distortion processes have been incorporated into some QA algorithms (such as those designed for the blocking artifact), the assumptions about the statistics of the i... |

274 |
Natural image statistics and neural representation,” Annu
- Simoncelli, Olshausen
- 2001
(Show Context)
Citation Context ... in [4]. Researchers believe that the visual stimulus emanating from the natural environment drove the evolution of the HVS, and that modeling natural scenes and the HVS are essentially dual problems =-=[5]-=-. While many aspects of the HVS have been studied and incorporated into quality assessment algorithms, a usefully comprehensive (and feasible) understanding is still lacking. NSS modeling may serve to... |

255 | The steerable pyramid: A flexible architecture for multi-scale derivative computation
- Simoncelli, Freeman
- 1995
(Show Context)
Citation Context ... −→ C i and −→ U i are M dimensional vectors, and we assume the RF U to be white. In this paper we model each subband of a scale-space-orientation wavelet decomposition (such as the steerable pyramid =-=[7]-=-) as a GSM. We partition the subband into non-overlapping blocks of M coefficients each, assume each block to be independent of others, and model each block as the vector −→ C i. With such a construct... |

212 | Image compression via joint statistical characterization in the wavelet domain
- Buccigrossi, Simoncelli
- 1999
(Show Context)
Citation Context ...d feasible) understanding is still lacking. NSS modeling may serve to fill this gap. NSS have been explicitly incorporated into a number of image processing algorithms: in compression algorithms [21]–=-=[24]-=-, denoising algorithms [25]–[27], image modeling [28], image segmentation [29], and texture analysis and synthesis [30]. While the characteristics of the distortion processes have been incorporated in... |

159 | Low-complexity image denoising based on statistical modeling of wavelet coefficients
- Mihcak, Kozintsev, et al.
- 1999
(Show Context)
Citation Context ...s still lacking. NSS modeling may serve to fill this gap. NSS have been explicitly incorporated into a number of image processing algorithms: in compression algorithms [21]–[24], denoising algorithms =-=[25]-=-–[27], image modeling [28], image segmentation [29], and texture analysis and synthesis [30]. While the characteristics of the distortion processes have been incorporated into some QA algorithms (such... |

148 |
Image quality measures and their performance
- Eskicioglu, Fisher
- 1995
(Show Context)
Citation Context ... include [7]–[14]. B. Arbitrary Signal Fidelity Criteria Researchers have also attempted to use arbitrary signal fidelity criteria in a hope that they would correlate well with perceptual quality. In =-=[15]-=- and [16], a number of these were evaluated for the purpose of QA. In [17], a structural similarity metric (SSIM) was proposed to capture the loss of image structure. SSIM was derived by considering h... |

141 | Bayesian Tree Structured Image Modeling Using Wavelet-Domain Hidden Markov Models - Romberg, Choi, et al. - 2001 |

140 | Perceptual image distortion - Teo, Heeger - 1994 |

125 | A visual discrimination model for imaging system design and evaluation - LUBIN - 1995 |

113 | On advances in statistical modeling of natural images
- Srivastava, Lee, et al.
- 2003
(Show Context)
Citation Context ...space of all possible signals. Many researchers have attempted to understand the structure of this subspace of natural images by studying their statistics. A good review on NSS models can be found in =-=[4]-=-. Researchers believe that the visual stimulus emanating from the natural environment drove the evolution of the HVS, and that modeling natural scenes and the HVS are essentially dual problems [5]. Wh... |

106 | Image quality assessment: From error measurement to structural similarity
- Wang, Bovik, et al.
(Show Context)
Citation Context ...ch researchers model the low-level features of the HVS to achieve consistent quality predictions. Although such methods have met with good success, there are many questions that arise in their design =-=[1]-=-. Some researchers have therefore also explored arbitrary signal fidelity criteria that are not affected by assumptions about HVS models, but are motivated instead by the need to capture the loss of v... |

104 | Modeling the joint statistics of images in the wavelet domain
- Simoncelli
- 1999
(Show Context)
Citation Context ...odeling. It is well known that the coefficients of a subband in a wavelet decomposition are neither independent nor identically distributed, though they may be approximately second-order uncorrelated =-=[32]-=-. A coefficient is likely to have a large variance if its neighborhood has a large variance. The marginal densities are sharply peaked around zero with heavy tails, which are typically modeled as Lapl... |

90 | A mathematical analysis of the DCT coefficient distributions for images - Lam, Goodman - 2000 |

89 | Random cascades on wavelet trees and their use in analyzing and modeling natural images
- Wainwright, Simoncelli, et al.
- 2001
(Show Context)
Citation Context ...nd feasible) understanding is still lacking. NSS modeling may serve to fill this gap. The natural scene model that we use in this paper is the Gaussian scale mixture (GSM) model in the wavelet domain =-=[6]-=-. A GSM is a random field (RF) that can be expressed as a product of two independent RFs. That is, a GSM C = { −→ C i : i ∈ I}, where I denotes the set of spatial indices for the RF, can be expressed ... |

88 | Multiscale image segmentation using wavelet-domain hidden markov models
- Choi, Choi, et al.
- 2001
(Show Context)
Citation Context ...s gap. NSS have been explicitly incorporated into a number of image processing algorithms: in compression algorithms [21]–[24], denoising algorithms [25]–[27], image modeling [28], image segmentation =-=[29]-=-, and texture analysis and synthesis [30]. While the characteristics of the distortion processes have been incorporated into some QA algorithms (such as those designed for the blocking artifact), the ... |

84 | Perceptual criteria for image quality evaluation
- Pappas, Safranek
- 2000
(Show Context)
Citation Context ...QA Methods HVS-based QA methods come in different flavors based on tradeoffs between accuracy in modeling the HVS and computational feasibility. A detailed discussion of these methods can be found in =-=[4]-=-–[6]. A number of HVS-based methods have been proposed in the literature. Some representative methods include [7]–[14]. B. Arbitrary Signal Fidelity Criteria Researchers have also attempted to use arb... |

67 | Issues in vision modeling for perceptual video quality assessment - Winkler - 1999 |

59 | An information fidelity criterion for image quality assessment using natural scene statistics
- Sheikh, Veciana
(Show Context)
Citation Context ...s to quantify the Shannon information shared between the test and the reference images, and showed that this shared image information is an aspect of fidelity that correlates well with visual quality =-=[3]-=-. In this paper we also quantify the information content of the reference image, since perceptual quality is likely to vary with the relative information loss, and propose a unified information fideli... |

59 | Statistical evaluation of image quality measures
- Avcbas, Sankur, et al.
- 2002
(Show Context)
Citation Context ...[7]–[14]. B. Arbitrary Signal Fidelity Criteria Researchers have also attempted to use arbitrary signal fidelity criteria in a hope that they would correlate well with perceptual quality. In [15] and =-=[16]-=-, a number of these were evaluated for the purpose of QA. In [17], a structural similarity metric (SSIM) was proposed to capture the loss of image structure. SSIM was derived by considering hypothetic... |

54 |
LIVE image quality assessment database Release 2
- Sheikh, Wang, et al.
- 2008
(Show Context)
Citation Context ...and finally a difference mean opinion score (DMOS) for each distorted image. The average RMSE for the DMOS was 5.92 with an average 95% confidence interval of width 5.48. The database is available in =-=[2]-=-. VI. RESULTS In this section, we present results on validation of VIF on the database presented in Section V, and present comparisons with other QA algorithms. Specifically, we compare the performanc... |

48 |
Perceptual quality metrics applied to still image compression
- Eckert, Bradley
- 1998
(Show Context)
Citation Context ...n the literature can be divided into two major groups: those based on the HVS and those based on arbitrary signal fidelity criteria (a detailed review of the research on FR QA methods can be found in =-=[3]-=-–[6]). A. HVS Error-Based QA Methods HVS-based QA methods come in different flavors based on tradeoffs between accuracy in modeling the HVS and computational feasibility. A detailed discussion of thes... |

42 | A wavelet visible difference predictor - Bradley - 1999 |

35 | Objective video quality assessment
- Wang, Sheikh, et al.
- 2003
(Show Context)
Citation Context ...sual structure in the signal that the HVS hypothetically extracts for cognitive understanding. Such top-down methods have also met with good success [1]. A review of recent QA methods can be found in =-=[2]-=-. In this paper we explore a novel information theoretic criterion for image fidelity using Natural Scene Statistics (NSS). Images and videos of the three dimensional visual environment come from a co... |

31 | DCTune: A technique for visual optimization of DCT quantization matrices for individual images - Watson - 1993 |

27 | A model of perceptual image fidelity - Heeger, Teo - 1995 |

26 | Image denoising using a local Gaussian scale mixture model in the wavelet domain
- Strela, Portilla, et al.
- 2000
(Show Context)
Citation Context ...articular realizations. The source model parameters that need to be estimated from the data consist of the field . For the vector GSM model, the maximum-likelihood estimate of can be found as follows =-=[36]-=-: (15) Estimation of the covariance matrix is also straightforward from the reference image wavelet coefficients [36] (16) In (15) and (16), is assumed to be unity without loss of generality [36]. B. ... |

25 | A haar wavelet approach to compressed image quality measurement - Lai, Kuo - 2000 |

21 | No-reference quality assessment using natural scene statistics: JPEG2000
- Sheikh, Bovik, et al.
(Show Context)
Citation Context ... the assumptions about the statistics of the images that they afflict are usually quite simplistic. Specifically, most QA algorithms assume that the input images are smooth and low-pass in nature. In =-=[31]-=-, an NSS model was used to design a no-reference image QA method for images distorted with the JPEG2000 compression artifacts. In this paper we use NSS models for FR QA, and model natural images in th... |

19 |
Quality assessment of coded images using numerical category scaling
- Dijk, Martens, et al.
- 1995
(Show Context)
Citation Context ...he reference images, were evaluated by human subjects in seven experiments. The raw scores were converted to difference scores (between the test and the reference) [37] and then converted to Z-scores =-=[38]-=-, scaled back to 1–100 range, and finally a difference mean opinion score (DMOS) for each distorted image. The average RMSE for the DMOS was 5.92 with an average 95% confidence interval of width 5.48.... |

16 |
Image quality metric based on multidimensional contrast perception models
- Pons, Malo, et al.
- 1999
(Show Context)
Citation Context ... computational feasibility. A detailed discussion of these methods can be found in [4]–[6]. A number of HVS-based methods have been proposed in the literature. Some representative methods include [7]–=-=[14]-=-. B. Arbitrary Signal Fidelity Criteria Researchers have also attempted to use arbitrary signal fidelity criteria in a hope that they would correlate well with perceptual quality. In [15] and [16], a ... |

3 |
Eero Simoncelli, “Image denoising using a local Gaussian Scale Mixture model
- Strela, Portilla
- 2000
(Show Context)
Citation Context ...tual information between the RF’s having the same statistics as those obtained from particular realizations. For the vector GSM model, the maximum-likelihood estimate of s 2 i can be found as follows =-=[9]-=-: �s 2 −→ T C i Cu i = −1−→ C i (15) M Estimation of the covariance matrix CU is also straightforward from the reference image wavelet coefficients [9]: �CU = 1 N N� i=1 −→ −→ T C i C i (16) In (15) a... |

3 |
et al., “Video quality experts group: Current results and future directions,” presented at
- Corriveau
- 2000
(Show Context)
Citation Context ... 982 images, out of which 203 were the reference images, were evaluated by human subjects in seven experiments. The raw scores were converted to difference scores (between the test and the reference) =-=[37]-=- and then converted to Z-scores [38], scaled back to 1–100 range, and finally a difference mean opinion score (DMOS) for each distorted image. The average RMSE for the DMOS was 5.92 with an average 95... |

2 |
JND-Metrix Technology. Evaluation Version available: http://www.sarnoff.com/products services/video vision/ jndmetrix/downloads.asp
- Corporation
- 2003
(Show Context)
Citation Context ...of the algorithm. 4. RESULTS In this section we present the results of our simulations using the VIF proposed in (14). We compare the performance of our algorithm against the well known Sarnoff model =-=[10]-=- and SSIM [1]. 4.1. Simulation Details For the wavelet decomposition, we used the steerable pyramid with six orientations [7]. These wavelets have better orientation selectivity, as well as a loose sh... |

1 | Image information and visual quality,” presented at the - Sheikh, Bovik - 2004 |