Multi-level Codebook Vector Quantization Method for Compression Coding of Hyperspectral Remote Sensing Images

Abstract

The invention discloses a multi-level codebook vector quantization method for compressing and encoding hyperspectral remote sensing images, and relates to the technical field of image processing. The method divides the spectral vector of the hyperspectral image into low-dimensional, medium-dimensional and high-dimensional according to the distortion situation. part, and then use the large-size codebook for the low-dimensional part with large distortion, use the medium-size codebook for the medium-dimensional part with little distortion, and use the small-size codebook for the high-dimensional part with small distortion, so that the multi-level codebook is adopted, Combined with the method of extracting only a quarter of the low-dimensional part to train the coding index after sorting the dispersion, it can effectively reduce the quantization distortion of the hyperspectral image and significantly reduce the calculation amount under the same compression ratio. The present invention can achieve higher-quality compression coding of hyperspectral images at a faster speed under the condition of less computational complexity, and has practical application value. It is a hyperspectral image lossy, near Lossless compression scheme.

Description

technical field [0001] The invention belongs to the field of hyperspectral remote sensing image processing, in particular to a multi-level codebook vector quantization scheme for compressing and encoding hyperspectral remote sensing images. Background technique [0002] In recent years, traditional two-dimensional color images have been far from meeting people's needs, and hyperspectral images have been replaced by more and more applications. A hyperspectral image is a spectral image formed by continuous spectral imaging of a target on hundreds of spectral bands by a hyperspectral imaging spectrometer with a nanoscale band width. It is defined as a three-dimensional image composed of a two-dimensional spatial domain and a one-dimensional spectral domain. Stereo data. This special image is different from ordinary two-dimensional grayscale images and color images, and it has the following characteristics: 1) Rich details and complex textures. Each ground target at the same s...

AI Technical Summary

Problems solved by technology

[0006] The present invention is aimed at various compression methods of existing hyperspectral images, or the compression quality is high but the calculation complexity is large, or the calculation amount is small but the image recovery quality is low, and its vector quantization scheme uses a unified codebook to quantify the components of each dimension, which lacks To solve the problem of data adaptability, a multi-level codebook vector quantization scheme for compressing and encoding hyperspectral remote sensing images is proposed

Images