Abstract:Discrete wavelet transform (DWT) provides an advantageous framework of multiresolution space-frequency representation with promising applications in image processing. The challenge and the opportunity in wavelet compression is the exploitation of the characteristics of the subband coefficients. In the decomposed subbands, the coefficients with the same magnitude generally do not have the same perceptual importance, depending on whether they belong to clustered scene structures. We propose an efficient scene adaptive and signal adaptive quantization scheme capable of exploiting, both the spectral and spatial localization properties resulting from wavelet transform. The proposed quantization is implemented as an MAP estimation based clustering process in which subband coefficients are quantized to their cluster means, subject to the spatial constraints enforced by a Markov-Gibbs random field. Consequently, the available bits are allocated to visually important scene structures so that the information loss is least perceptible by the human viewers. This adaptive quantization represents an effort to develop compression algorithms for low bit rates through the exploitation of both the information theory and the human visual system. It is fundamentally different from the existing scalar and vector quantization schemes in that only the proposed adaptive quantization can differentiate the perceptual importance of the subband coefficients. The concept and the mechanism of the scene structure preservation are applicable and tunable to other image processing tasks. This research also investigates several related topics, including a zerotree-based coding scheme to exploit the dependencies of coefficients across different wavelet scales, a face location and region-based prioritized coding for high perceptual quality videoconferencing, and a generic image enhancement technique for the reduction of coding artifacts at low bit rates. In addition, perspectives of a wavelet-based compression scheme for wireless communications is analyzed. Experimental results in various applications have shown the potential of the proposed scheme for low bit rate image and video coding.

A New and Efficient Audio Compression Based on Ebcot and Rvlc

An Improved Wavelet Coefficients Symbols Coding of Embedded Zero-tree Wavelet

An Audio Compression Algorithm Using a Wavelet Packet Transform

A Audio Coding Algorithm for Interactive Communications

An Intra-BRNN and GB-RVQ Based END-TO-END Neural Audio Codec

A Novel Embedded Coding Algorithm Based on the Reconstructed DCT Coefficients.

VRVQ: Variable Bitrate Residual Vector Quantization for Audio Compression

ERVQ: Enhanced Residual Vector Quantization with Intra-and-Inter-Codebook Optimization for Neural Audio Codecs

Lifting Wavelet Transform Based on Audio Lossless Coding Method

Improved HEVC Video Compression Algorithm Using Low-Complexity Frame Rate Up Conversion

Improved Image Compression Method Based on Wavelet Transform and Its Application

A Predictive VQ Based Video Compression Scheme

High-Fidelity Audio Compression with Improved RVQGAN

A Scalable Digital Audio Encoder Based on Embedded Zerotree Wavelet

Zerotree Wavelet Image Coder with Quantization

Low-bit-rate wavelet-based image compression using human visual characteristics

A Study of the Improved Algorithm of ROI Image Compression Based on Wavelet Transform

High Efficiency Vibrotactile Codec Based on Gate Recurrent Network

Efficient Vibrotactile Codec Based on Nbeats Network

Low Bit Rate Wavelet-Based Image and Video Compression with Adaptive Quantization, Coding and Postprocessing

High Efficiency RVLC Construction Algorithm