Quantization (image processing)

About: Quantization (image processing) is a research topic. Over the lifetime, 7977 publications have been published within this topic receiving 126632 citations.

Error resilient pyramid vector quantization for image compression

Abstract: The robustness of image and video compression in the presence of time-varying channel error has received increased interest with the emergence of portable digital receivers and computers. To achieve robust compression, pyramid vector quantization (PVQ) can be used. It is a fixed-rate quantization scheme suited to Laplacian-like sources, such as those arising from transform and subband image coding. The authors propose a new method of PVQ which not only improves upon the image compression performance of typical JPEG implementations, but also demonstrates excellent resilience to channel error. >

The lossless DCT for combined lossy/lossless image coding

Abstract: This paper introduces a modification of the discrete cosine transform (DCT) that produces integer coefficients from which the original image data can be reconstructed losslessly. It describes an embedded coding scheme which incorporates this lossless DCT and presents some experimental rate-distortion curves for this scheme. The results show that the lossless compression ratio of the proposed scheme exceeds that of the lossless JPEG predictive coding scheme. On the other hand, in lossy operation the rate-distortion curve of the proposed scheme is very close to that of lossy JPEG. Also, the transform coefficients of the proposed scheme can be decoded with the ordinary DCT at the expense of a small error, which is only significant in lossless operation.

Optimal Gradient Compression for Distributed and Federated Learning.

Abstract: Communicating information, like gradient vectors, between computing nodes in distributed and federated learning is typically an unavoidable burden, resulting in scalability issues. Indeed, communication might be slow and costly. Recent advances in communication-efficient training algorithms have reduced this bottleneck by using compression techniques, in the form of sparsification, quantization, or low-rank approximation. Since compression is a lossy, or inexact, process, the iteration complexity is typically worsened; but the total communication complexity can improve significantly, possibly leading to large computation time savings. In this paper, we investigate the fundamental trade-off between the number of bits needed to encode compressed vectors and the compression error. We perform both worst-case and average-case analysis, providing tight lower bounds. In the worst-case analysis, we introduce an efficient compression operator, Sparse Dithering, which is very close to the lower bound. In the average-case analysis, we design a simple compression operator, Spherical Compression, which naturally achieves the lower bound. Thus, our new compression schemes significantly outperform the state of the art. We conduct numerical experiments to illustrate this improvement.

JPEG 2000 next generation image compression system features and syntax

Abstract: As the resolution and pixel fidelity of digital imagery grows, there is a greater need for more efficient compression and extraction of images and sub-images. The ability to handle many types of image data, extract images at different resolutions and quality, lossless and lossy, zoom and pan, and extract regions-of-interest is the new measures of image compression system performance. JPEG 2000 is designed to address the needs of high quality imagery. This paper describes how the JPEG 2000 syntax and file format support these features. The decomposition of the image into the codestream is described along with associated syntax markers. Examples of how the syntax enables some of the features of JPEG 2000 are offered.

Method of video encoding by accumulated error processing and encoder therefor

Abstract: The present invention relating to a method of video coding associated with processing accumulated errors and a encoder therefor, the method comprising the steps of: (a) generating motion vectors of an input image in a predetermined unit and the difference image between an image of filtering a motion-compensated image on a reconstructed previous frame and the input image on current frame, and then performing discrete cosine transform (DCT), quantization and variable length coding on the difference image; (b) generating the motion-compensated image on the reconstructed previous frame from the reconstructed previous frame and the motion vectors; and (c) filtering off accumulated errors while preserving the edges within the motion-compensated image on the reconstructed previous frame. Therefore, random distributed noises due to accumulated errors can be removed and bit generation amounts by filtering off random accumulated errors with a high frequency characteristics before coding can be reduced.