An efficient image coder based on the combination of visual and probabilistic models

Document(s) associé(s)

Annexes

• AS_14_n2_137_CHE.pdf (2,9M)

Abstract

To maintain high image quality with low bitrate, an effective coding algorithm should not only remove statistical correlation but also perceptual redundancy from image signals. In this paper, a coding scheme based on statistical and perceptual models is presented. Roughly speaking, the compression process is perfonned in two stages. In the first stage we derive a perceptual measure called the just noticeable distortion (JND). This measure provides each signal being coded with an error visibility threshold, below which reconstruction error are rendered imperceptible. Using this threshold the perceptually insignificant signals can be screened out. In the second stage, we perform a quantization thresholding step using statistical information on the DCT coefficient distribution. Such that statistical redundancy is minimized. Using this system, high quality is obtainable at rates lower than 1 bit/pixel for a wide variety of images without user intervention.

Pour citer cet article

A propos de : Hocine Cherifi

A propos de : Thierry Eude