Hyperspectral Image Coding Using Spectral Prediction Modelling And Optimal Compression Plane

A typical visible light spectrum for a human is about 400 nanometres (nm) to 700 nm in wavelength. Hyperspectral (HS) images include hundreds of narrow and contiguous spectral bands that cover a wide range of wavelengths extending from 400 to 2500 nm. A number of applications based on HS images are rapidly expanding in different areas, including environmental monitoring, biological and medical analysis assistance, agricultural surveillance, mineralogy mapping, and food quality analysis, etc.
HS images spectral signatures can provide moisture content, texture, reflectance and other external quality characteristics of diverse samples far beyond human vision. This ability comes at a price: big data and high redundancy. Thus, exploring a compression technique that is effective against multidimensional data and different from traditional compression algorithms is vital.
A directionlet based compression scheme and constituted the optimal compression plane (OCP) for adaptive best approximation of the geometric matrix were developed. The OCP, calculated by the spectral correlation, is used to predict and determine which reconstructed plane can reach higher compression rates while minimising data loss of hyperspectral data. Furthermore, a fundamental shift is required from the existing “original pixel intensity”-based coding approaches using traditional image coders (e.g. JPEG2000) to the “residual” based approaches using a video coder for better compression performance. A novel coding framework using Reflectance Prediction Modelling (RPM) in the latest video coding standard High Efficiency Video Coding (HEVC) for HS images is proposed. The modelling can predict the distribution and correlation of the pixel vectors for different bands. Every spectral band of a HS image is treated as if it is an individual frame of a video. A HS image presents a wealth of data where every pixel is considered as a vector for each spectral band. The pixel vector’s distribution along spectral bands can be determined by quantitative comparison and analysis. Therefore The conclusion is that the Multilayer Propagation Neural Networks (MLPNN) with back propagation training are particularly well suited for addressing the approximation function.
A block-based inter-band predictor (BIP) with a multilayer propagation neural network model (BIP-MLPNN) aims to further reduce any intra-band residual and a MLPNN model hybrid with Adaptive Prediction Mechanism (APM) is proposed. BIP-MLPNN is trained to predict the succeeding bands by using current band information to explore whether BIP-MLPNN can provide better image compression results in HS images and to obtain further compression results. The experimental results are fully justified by three types of HS datasets with different wavelength ranges. The proposed method outperforms the existing mainstream HS encoders in terms of rate-distortion performance of HS image compression.