Hyperspectral Image Compression Research Articles

Hybrid hyperspectral image compression technique for non-iterative factorized tensor decomposition and principal component analysis: application for NASA’s AVIRIS data

Remote sensing data has shown tremendous potential for applications in various fields like land use mapping and detection, geologic mapping, water resource applications, wetland mapping, urban and regional planning, environment inventory, natural disaster assessment, archaeological applications, and others. Every day, thousands of gigabytes of memory are involved in capturing the hyperspectral remote sensing datasets. The compelling information present in these hyperspectral images (HSIs) is very minimal due to redundancy. Spatial and spectral correlations monopolize the acquired HSI data sets. Therefore, an algorithm that exploits these correlations and compresses the HSI tensors is proposed in this paper. First, the acquired HSI image (reflectance data) is subjected to the removal of geometric and radiometric errors. Second, spectral bands of interest affiliated to the underlying application are exclusively processed for principal component analysis (PCA). Results of this PCA are scrutinized to identify the absolute dependent components. Further, these components are exposed to a non-iterative factorized compression technique. As a result, HSI 3D tensors are disintegrated into 1D tensors. This tensor breakdown leads to a compression ratio as high as 3747:1 while the total encoding time observed is 332 s and RMSE is as low as 0.0017. Later, the original HSI is reconstructed back by the product of decomposed individual tensors and its PSNR is 53.03 dB. The proposed compression method targets the tucker decomposition-based HSI compression technique which is computationally complex and time consuming, and hence, a breakthrough is achieved with the technique introduced.

Performance Impact of Parameter Tuning on the CCSDS-123.0-B-2 Low-Complexity Lossless and Near-Lossless Multispectral and Hyperspectral Image Compression Standard

This article studies the performance impact related to different parameter choices for the new CCSDS-123.0-B-2 Low-Complexity Lossless and Near-Lossless Multispectral and Hyperspectral Image Compression standard. This standard supersedes CCSDS-123.0-B-1 and extends it by incorporating a new near-lossless compression capability, as well as other new features. This article studies the coding performance impact of different choices for the principal parameters of the new extensions, in addition to reviewing related parameter choices for existing features. Experimental results include data from 16 different instruments with varying detector types, image dimensions, number of spectral bands, bit depth, level of noise, level of calibration, and other image characteristics. Guidelines are provided on how to adjust the parameters in relation to their coding performance impact.

Hyperspectral image compression using hybrid transform with different wavelet-based transform coding

Hyperspectral image resolution offers limited spectral bands within a continual spectral spectrum, creating one of the spectra of most pixels inside the sequence which contains huge volume of data. Data transmission and storage is a challenging task. Compression of hyperspectral images are inevitable. This work proposes a Hyperspectral Image (HSI) compression using Hybrid Transform. First the HSI is decomposed into 1D and it is clustered and tiled. Each cluster is applied with Integer Karhunen–Loeve Transform (IKLT) and as such it is applied for whole image to get IKLT bands in spectral dimension. Then IKLT bands are applied with Integer Wavelet Transform (IDWT) to decorrelate the spatial data in spatial dimension. The combination of IKLT and IDWT is known as Hybrid transform. Second, the decorrelated wavelet coefficients are applied to Spatial-oriention Tree Wavelet (STW), Wavelet Difference Reduction (WDR) and Adaptively Scanned Wavelet Difference Reduction (ASWDR). The experimental result shows STW algorithm using Hybrid Transform gives better PSNR (db) and bits per pixel per band (bpppb) for hyperspectral images. The comparison between STW, WDR and ASWDR with Hybrid Transform for Indian Pines, Salinas, Botswana, Botswana and KSC images is experimented.

Complex analysis and monitoring of the environment based on earth sensing data

We study a problem of complex analysis and monitoring of the environment based on Earth Sensing Data, with the emphasis on the use of hyperspectral images (HSI), and propose a solution based on developing algorithmic procedures for HSI processing and storage. HSI is considered as a two-dimensional field of pixel signatures. Methods are proposed for evaluating the similarity of a HSI pixel signature with a reference signature, via simple alignment transformations: identical; amplitude scaling; shift along y-axis; and a combination of the last two. A clustering / recognition method with self-learning is proposed, which determines values of the transformation parameters that ensure the alignment of the current pixel signature with the reference signature. Similarity with the reference is determined by a standard deviation value. A HSI compression method with controlled losses has been proposed. The method forms a basis via accumulating reference signatures and represents the rest of the signatures by parameters matching them with the already detected class-reference signature. In an experiment with the GSI f100520t01p00-12 data of the AVIRIS spectrometer, the method provided a 2 % loss and compression coefficients of the original HSI ranging from 43 to 165 for various types of alignment transformation, while not requiring archiving and thus maintaining active access to the HSI and using the list of references as an analogue of the HSI palette. An algorithm for the formation of dense groups of detectable objects (for example, oil spots) and their nonconvex contouring, controlled by 4 parameters, is proposed. A pilot version of the concept of geographic information system (GIS) and an appropriate database management system (DBMS) was built and implemented, which provides monitoring and is based on the prioritized processing and storage of the HSI, which serve as a data source for the system. A laboratory complex with new algorithms for processing and storing the GSE is introduced into the structure of the system.

Lossless compression for hyperspectral image using deep recurrent neural networks

With the rapid development of hyperspectral remote sensing technology, the spatial resolution and spectral resolution of hyperspectral images are continually increasing, resulting in a continual increase in the scale of hyperspectral data. At present, hyperspectral lossless compression technology has reached a bottleneck. Simultaneously, the rise of deep learning has provided us with new ideas. Therefore, this paper examines the use of deep learning for the lossless compression of hyperspectral images. In view of the differential pulse code modulation (DPCM) method being insufficient for predicting spectral band information, the proposed method, called C-DPCM-RNN, uses a deep recurrent neural network (RNN) to improve the traditional DPCM method and improve the generalization ability and prediction accuracy of the model. The final experimental result shows that C-DPCM-RNN achieves better compression on a set of calibrated AVIRIS test images provided by the Multispectral and Hyperspectral Data Compression Working Group of the Consultative Committee for Space Data Systems in 2006. C-DPCM-RNN overcomes the limits of traditional methods in its performance on uncalibrated AVIRIS test images.

New Results in Perceptually Lossless Compression of Hyperspectral Images

Hyperspectral images (HSI) have hundreds of bands, which impose heavy burden on data storage and transmission bandwidth. Quite a few compression techniques have been explored for HSI in the past decades. One high performing technique is the combination of principal component analysis (PCA) and JPEG-2000 (J2K). However, since there are several new compression codecs developed after J2K in the past 15 years, it is worthwhile to revisit this research area and investigate if there are better techniques for HSI compression. In this paper, we present some new results in HSI compression. We aim at perceptually lossless compression of HSI. Perceptually lossless means that the decompressed HSI data cube has a performance metric near 40 dBs in terms of peak-signal-to-noise ratio (PSNR) or human visual system (HVS) based metrics. The key idea is to compare several combinations of PCA and video/ image codecs. Three representative HSI data cubes were used in our studies. Four video/image codecs, including J2K, X264, X265, and Daala, have been investigated and four performance metrics were used in our comparative studies. Moreover, some alternative techniques such as video, split band, and PCA only approaches were also compared. It was observed that the combination of PCA and X264 yielded the best performance in terms of compression performance and computational complexity. In some cases, the PCA + X264 combination achieved more than 3 dBs than the PCA + J2K combination.

Lossy Compressive Sensing Based on Online Dictionary Learning

In this paper, a lossy compression of hyperspectral images is realized by using a novel online dictionary learning method in which three dimensional datasets can be compressed. This online dictionary learning method and blind compressive sensing (BCS) algorithm are combined in a hybrid lossy compression framework for the first time in the literature. According to the experimental results, BCS algorithm has the best compression performance when the compression bit rate is higher than or equal to 0.5 bps. Apart from observing rate-distortion performance, anomaly detection performance is also tested on the reconstructed images to measure the information preservation performance.

The linear prediction vector quantization for hyperspectral image compression

In this paper, a hyperspectral image compression method is proposed. It is based on spectral clustering, linear prediction and the vector quantization (VQ). Since the hyperspectral image has stronger spectral correlation than spatial correlation, the spectral clustering and model of linear prediction are introduced to reduce the spectral correlation. In the proposed method, spectral clustering algorithm of K-means is employed, and the centroids of clustered results are used as reference bands, then the reference bands are employed in the model of linear prediction to compute the prediction error, finally the prediction error is encoded by VQ. The experiments results using AVIRIS images are compared to IVQ and AR + SubPCA+JPEG2000 algorithm, the results show that our proposed algorithm outperforms other algorithms.

Universal Golomb–Rice Coding Parameter Estimation Using Deep Belief Networks for Hyperspectral Image Compression

For efficient compression of hyperspectral images, we propose a universal Golomb–Rice coding parameter estimation method using deep belief network, which does not rely on any assumption on the distribution of the input data. We formulate the problem of selecting the best coding parameter for a given input sequence as a supervised pattern classification problem. Simulations on the synthesized data and five hyperspectral image datasets show that we can achieve significantly more accurate estimation of the coding parameters, which can translate to slightly higher compression than three state-of-the-art methods. More extensive simulations on additional images from the 2006 AVIRIS datasets show that the proposed method achieved overall compression bitrates comparable with other estimation methods, as well as the sample-adaptive entropy coder employed by the Consultative Committee for Space Data Systems standard for multispectral and hyperspectral data compression. Regarding computational feasibility, we show how to use transferable deep belief networks to speed up training by about five times. We also show that inferring the best coding parameters using a trained deep belief network offers computational advantages over the brute-force search method. As an extension, we propose a novel side-information free codec, where the intersequence correlations can be learned by a differently trained network based on the current sequence to predict reasonably good parameters for coding the next sequence. As another extension, we introduce a variable feature combination architecture, where problem specific heuristics such as the sample means can be incorporated to further improve the estimation accuracy.

Hyperspectral image compression based on simultaneous sparse representation and general-pixels

Simultaneous sparse representation can transform the correlated spectral signatures of hyperspectral pixel matrixes into sparse coefficients. It can be very efficient in the compression scheme when the original image is clustered to general-pixels (a cluster of hyperspectral pixels which contains the similar signature). In this paper, we propose a simultaneous sparse representation based hyperspectral image compression scheme. First, the whole hyperspectral pixels are clustered into general-pixels and each general-pixel will be coded by the simultaneous sparse representation scheme. To further compress the coefficients, the differential pulse code modulation filter is adopted in each row coefficients. Finally, all the nonzero coefficients, over-complete dictionary and mapping data of general-pixels will be transformed into the binary bitstream by Huffman coding. The results on four hyperspectral image datasets show that our method outperforms several classical and the state-of-the-art methods in term of rate-distortion and spectral fidelity performance.

Superpixel based recursive least-squares method for lossless compression of hyperspectral images

Filtering based compression methods have become a popular research topic in lossless compression of hyperspectral images. Recursive least squares (RLS) based prediction methods provide better decorrelation performance among the filtering based methods. In this paper, two superpixel segmentation based RLS methods, namely SuperRLS and B-SuperRLS, are investigated for lossless compression of hyperspectral images. The proposed methods present a novel parallelization approach for RLS based prediction method. In the first step of SuperRLS, superpixel segmentation is applied to hyperspectral image. Afterwards, hyperspectral image is partitioned into multiple small regions according to the superpixel boundaries. Each region is predicted with RLS method in parallel, and prediction residuals are encoded via arithmetic encoder. Additionally, superpixel based prediction approach provides region of interest compression capability. B-SuperRLS, which is bimodal version of SuperRLS, evaluates both spectral and spatio-spectral correlations for prediction. The performance of the proposed methods are exhaustively analysed in terms of superpixel number, input vector length and number of parallel nodes, used in the prediction. Experimental results show that the proposed parallel architecture dramatically reduces the computation time, and achieves lower bit-rate performances among the state-of-the-art methods.

Compressed Sensing Reconstruction of Hyperspectral Images Based on Spectral Unmixing

How to utilize the characteristics of hyperspectral images (HSIs) is a key problem in application of compressed sensing theory to hyperspectral image compression and reconstruction. Based on the study of spectral mixing characteristics, a compressed sensing reconstruction algorithm with spectral unmixing for HSIs is proposed. Taking advantage of linear mixing model, the HSIs are separated into endmember matrix and abundance matrix. Instead of directly reconstructing the entire hyperspectral data as traditional reconstruction algorithms, the proposed algorithm explores the idea of spectral unmixing for reconstruction. In the sampling process, the HSIs are sampled both spatially and spectrally. In the reconstruction process, a joint optimization problem for endmember extraction and abundance estimation is established and solved in an iterative way to obtain the reconstructed hyperspectral data. Experimental results on synthetic and real hyperspectral data demonstrate that the proposed algorithm could obtain the endmember and abundance information effectively, and the accuracy of reconstructed HSIs as well as the computational efficiency are superior to the state-of-the-art reconstruction algorithms.

Large-scale hyperspectral image compression via sparse representations based on online learning

Abstract In this study, proximity based optimization algorithms are used for lossy compression of hyperspectral images that are inherently large scale. This is the first time that such proximity based optimization algorithms are implemented with an online dictionary learning method. Compression performances are compared with the one obtained by various sparse representation algorithms. As a result, proximity based optimization algorithms are listed among the three best ones in terms of compression performance values for all hyperspectral images. Additionally, the applicability of anomaly detection is tested on the reconstructed images.

The Hyper-spectral Image Compression Based on K-Means Clustering and Parallel Prediction Algorithm*

In this paper, we propose a lossless compression algorithm for hyper-spectral images with the help of the K-Means clustering and parallel prediction. We use K-Means clustering algorithm to classify hyper-spectral images, and we obtain a number of two dimensional sub images. We use the adaptive prediction compression algorithm based on the absolute ratio to compress the two dimensional sub images. The traditional prediction algorithm is adopted in the serial processing mode, and the processing time is long. So we improve the efficiency of the parallel prediction compression algorithm, to meet the needs of the rapid compression. In this paper, a variety of hyper-spectral image compression algorithms are compared with the proposed method. The experimental results show that the proposed algorithm can effectively improve the compression ratio of hyper-spectral images and reduce the compression time effectively.

动态测量的高光谱图像压缩感知

动态测量的高光谱图像压缩感知

Compression of hyper-spectral images using an accelerated nonnegative tensor decomposition

Abstract Nonnegative tensor Tucker decomposition (NTD) in a transform domain (e.g., 2D-DWT, etc) has been used in the compression of hyper-spectral images because it can remove redundancies between spectrum bands and also exploit spatial correlations of each band. However, the use of a NTD has a very high computational cost. In this paper, we propose a low complexity NTD-based compression method of hyper-spectral images. This method is based on a pair-wise multilevel grouping approach for the NTD to overcome its high computational cost. The proposed method has a low complexity under a slight decrease of the coding performance compared to conventional NTD. We experimentally confirm this method, which indicates that this method has the less processing time and keeps a better coding performance than the case that the NTD is not used. The proposed approach has a potential application in the loss compression of hyper-spectral or multi-spectral images

Hyperspectral image compression approaches: opportunities, challenges, and future directions: discussion.

This paper establishes a review of the recent study in the field of hyperspectral (HS) image compression approaches. Recently, image compression techniques have achieved significant advances from diverse types of coding standards/approaches. HS image compression requires an unconventional coding technique because of its unique, multiple-dimensional structure. The data redundancy exists in both inter-band and intra-band methods. The survey summarizes current literature in inter- and intra-band compression methods. The challenges, opportunities, and future research possibilities regarding HS image compression are further discussed. The experimental results are also provided for validity and applicability of the existing HS image compression techniques.

Lossless hyperspectral image compression using bimodal conventional recursive least-squares

ABSTRACTHyperspectral image compression is an important task, where the aim is to store or transmit data in an efficient way. Hyperspectral images are mostly captured by a sensor system that includes multiple imaging sensors covering different regions of the electromagnetic spectrum. Misalignment of multiple imaging sensors produces boresight effect, and this problem can degrade band prediction performance noticeably. Another problem is prediction of blurry band images. In order to gain robustness to these problems, bimodal conventional recursive least-squares (B-CRLS) prediction method is proposed for the lossless compression of hyperspectral images. Two prediction modes are defined: spectral and spatio-spectral. B-CRLS method has a two-step process. First, mode selection is carried out for each band. Afterwards, final band image is predicted by using the selected mode, and the residual images are encoded with an arithmetic encoder. The proposed method is compared to adaptive-length CRLS, fixed-length CRLS, and other well-known prediction methods. Experiments have been performed on uncalibrated and calibrated hyperspectral images. Obtained results show that the proposed method achieves competitive compression performance with the state-of-the-art while providing relatively lower-computation times.

Parallel Implementation of the Recursive Least Square for Hyperspectral Image Compression on GPUs

Compression is a very important technique for remotely sensed hyperspectral images. The lossless compression based on the recursive least square (RLS), which eliminates hyperspectral images` redundancy using both spatial and spectral correlations, is an extremely powerful tool for this purpose, but the relatively high computational complexity limits its application to time-critical scenarios. In order to improve the computational efficiency of the algorithm, we optimize its serial version and develop a new parallel implementation on graphics processing units (GPUs). Namely, an optimized recursive least square based on optimal number of prediction bands is introduced firstly. Then we use this approach as a case study to illustrate the advantages and potential challenges of applying GPU parallel optimization principles to the considered problem. The proposed parallel method properly exploits the low-level architecture of GPUs and has been carried out using the compute unified device architecture (CUDA). The GPU parallel implementation is compared with the serial implementation on CPU. Experimental results indicate remarkable acceleration factors and real-time performance, while retaining exactly the same bit rate with regard to the serial version of the compressor.

Progressive lossy-to-lossless coding of hyperspectral images through regression wavelet analysis

ABSTRACTProgressive Lossy-to-Lossless (PLL) coding techniques enable a gradual quality improvement of the recovered images, starting from a coarse approximation up to a perfect reconstruction. PLL is becoming a widespread approach in several scenarios, in particular, for compression of hyperspectral images. In this paper we assess the suitability of Regression Wavelet Analysis (RWA) for hyperspectral image progressive lossy-to-lossless coding. RWA is a recent spectral transform that combines a wavelet transform with a regression stage, providing excellent coding performance for lossless compression. When coupled with a pyramidal predictive weighting scheme, RWA also yields very competitive coding results for PLL at a low computational cost. Coding performance is assessed within the framework of Joint Photographic Experts Group (JPEG) 2000 standard, comparing RWA against state-of-the-art spectral transforms, including reversible Karhunen-Loève Transform (rKLT) and Pairwise Orthogonal Transform (POT). Comparison with respect to Multiband Context-based Adaptive Lossless/Near-Lossless Image Coding (M-CALIC) technique is also provided. Experiments are conducted on uncalibrated and calibrated hyperspectral images from Airborne Visible/Infrared Imaging Spectrometer (AVIRIS), satellite-borne Hyperion and Infrared Atmospheric Sounding Interferometer (IASI) sensors. Discussion embraces rate-distortion performance, bit-per-pixel-per-component rate distribution and classification outcome.