<title>Design of psychovisual quantizers for a visual subband image coding</title>

Abstract

In order to specify the optimal psychovisual quantizer associated to a given visual subband decomposition scheme [1], a new methodology has been developped. Psychovisual experiments based on the visibility of the quantization noise, have been conducted. The complex signals used, have been caracterised by the local bandlimited contrast. The main finding of this study is that the quantizers are, in the chosen space contrast ,of a linear type. The quantization intervals contrast, obtained with a given observer, are of 0.039 for the subband called III-1 (radial selectivity 1 .5cy/d° to 5.7cy/d°, angular selectivity -22.5d° to 22.5) 0.03 1 for the subband called IV-1 (radial selectivity 5.7cy/d° to 14. lcy/d°, angular selectivity -15d°to 1 Sd°) 0.1 17 for the subband called V-i (radial selectivity 14.icy/d° to 28.2cy/d°, angular selectivity -i5d° to 15d°) To evaluate the importance of the "angular" aspect in this approach, further measurements have been made with the subband IV -2 (radial selectivity 5.7cy/d° to 14. icy/d°, angular selectivity 15d° to 30d°). The linearity is also observed and the quantization interval contrast, for the same observer, is of 0.122.Key words: Psychovisual quantizer,visual subband, Local band-limited contrast, Visibility of quantization noise 1 . INTRODUCTIONThe design of quantizers involves specifying thresholds and levels which minimize both entropy and some measure of distorsion. Whereas for the "classical" quantizers, such distorsion may be well measured by the rms (root mean square) error, for the psychovisual quantizers we are concerned by the visual distorsion. For the human visual system, the same error is less detectable when it is superimposed with strong signals (with high values ) than when it is superimposed with weak ones ( small values).To design a psychovisual quantizer, A.B.Watson [2] used the contrast masking function. This function represents, for a given observer, the discrimination between two images that differ only in contrast. In his experiments the first image (the background) has a contrast c and the second one (the background plus increment) has a contrast c + L\c. The increment discrimination thresholds measured have then been modeled bywhere Ec(O) represents the contrast detection threshold when the background contrast is zero. This function has been applied directly to the design of the quantizer. The quantization thresholds T and the reconstruction levels L1 have been determined by 1446 ISPIE Vol. 2308 0-8194-1638-X/94/$6.00 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

MPEG-7 Homogeneous Texture Descriptor

“…Eight bits are assigned and used for linear quantization of each descriptor value. The linear quantization used in the paper is written as (11) where D quant is a quantized descriptor value, D nonquant is the value of the feature, and max β and min β are the maximum and minimum values of features in the database. Note that q_level is set to 255.…”

“…It provided higher retrieval accuracy for the testing data sets. Therefore, the homogeneous texture descriptor described in this paper was selected as the normative MPEG-7 homogeneous texture descriptor in the Visual part of the MPEG-7 final committee draft [11]- [14].…”

<title>Vector quantization based on a psychovisual lattice for a visual subband coding scheme</title>

“…In previous study [8], psychophysical experiments have been performed to evaluate quantization impairment perception (at normalized viewing distance of 4H) upon natural pictures. In the local band limited contrast space it was shown that contrast must always be uniformly quantized in order to achieve a just noticeable quantization noise (the quantization step Aj being dependent of the visual subband considered).…”

<title>Interactions of chromatic components in the perceptual quantization of the achromatic component</title>

“…It is more convenient here to study the visibility produced by quantization of the content of a particularly subband rather than the visibility of any increments or any white gaussian noises. Previously, we have shown 5 that perception of quantization noise on L ij at location (m,n) is directly dependent on the ratio between L ij and the average luminance at this location. This latter is computed from the subbands having a lower radial frequency.…”

“…Here we are concerned with Krauskopf's directions A, Cr1, Cr2. For each components, it can be identified both analysis and quantization stages like we have done in previous works 4,5 . Unfortunately, perceptual components are not independent and some interactions subsist between luminance and chrominance.…”