Linear transform for simultaneous diagonalization of covariance and perceptual metric matrix in image coding

Epifanio, Irene; Gutiérrez, Juan Manuel; Malo, Jesús

doi:10.1016/s0031-3203(02)00325-4

Cited by 33 publications

(63 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The perceptual dependence may be described by a non-Euclidean perceptual metric, W [2]. The efficiency of a representation from both points of view may be evaluated analyzing the non-diagonal nature of these matrices.…”

Section: Codingmentioning

confidence: 99%

“…2. As this slope is non-diagonal and input-dependent, the metric, W , cannot be diagonalized in any linear domain (see [1], [2] for details).…”

Section: Codingmentioning

confidence: 99%

“…The aim of the change of representation in transform coding is twofold [2]: it is intended to remove the statistical and the perceptual dependence between the image coefficients.…”

Section: Introductionmentioning

confidence: 99%

“…Moreover, in order to apply the standard theory for bit allocation, the (perceptual) metric has to be diagonal in the representation to be quantized [4]. However, the above linear transforms do not completely achieve the desired independence from both points of view [2], [6]. This means that scalar quantization of these representations is not completely appropriate.…”

Section: Introductionmentioning

confidence: 99%

“…This divisive normalization non-linearity is the transform that takes place after the local-frequency analysis in biological early vision [8], [9]. Besides, this kind of perceptually inspired normalization naturally leads to a perceptually Euclidean domain [2], [10].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Normalized image representation for efficient coding

Malo

The Thrity-Seventh Asilomar Conference on Signals, Systems &Amp; Computers, 2003

Self Cite

View full text Add to dashboard Cite

Abstract-In this paper we propose an adaptive non-linear image representation based on the divisive normalization of localfrequency transforms used in contrast masking models. This normalized representation has two effects: (1) it increases the statistical independence of the coefficients of the representation and (2) it is Euclidean from a perceptual point of view.Experimental results show that reducing the remaining statistical and perceptual dependence using normalized representations for transform coding may make a big difference in the quality of the reconstructed images.

show abstract

Section: Codingmentioning

confidence: 99%

“…2. As this slope is non-diagonal and input-dependent, the metric, W , cannot be diagonalized in any linear domain (see [1], [2] for details).…”

Section: Codingmentioning

confidence: 99%

“…The aim of the change of representation in transform coding is twofold [2]: it is intended to remove the statistical and the perceptual dependence between the image coefficients.…”

Section: Introductionmentioning

confidence: 99%