Multimodal Compression Research Articles

Compressing Biosignal for Diagnosing Chronic Diseases

Electrocardiogram, electromyogram, electroencephalogram are the foremost required vital signs for diagnosing chronic diseases like sleep disorder, mood disorder, epilepsy etc., which demands long-term monitoring. A sensor based wearable system which is enabled with internet technology, supports the continuous recordings of these vital signs without troubling the patient’s daily activities. And the wearable hub is responsible for collecting the readings of bio signals from multiple micro-sensor nodes deployed around the body which creates the short range of communication and forward to the observer. These continuous monitoring increases the signal transmission cost and declines the battery life of wearables. So, the observed multiple bio signals can be compressed jointly than individually before sending, at an edge level. This paper proposes transfer learning based multimodal convolutional denoising autoencoder to perform multimodal compression and to reconstruct the data from its latent representation. Transfer learning helps the system to reuse the learned weights which may reconstruct the data with better quality score than by randomly initialized weights. The proposed work achieves compression ratio of 128 and it is proved that multimodal compression is better than unimodal compression in case of consuming multiple sensors. And the experimental result proves that the computation cost is low in multimodal compression than in unimodal compression.

An efficient JPEG-2000 based multimodal compression scheme

In this paper, a wavelet-based multimodal compression method is proposed. The method jointly compresses a medical image and an ECG signal within a single codec, i.e., JPEG-2000 in an effective and simple way. The multimodal scheme operates in two main stages: the first stage, consists of the encoder and involves a mixing function, aiming at inserting the samples of the signal in the image according to a predefined insertion pattern in the wavelet domain. The second stage represented by a separation function, consists of the extraction process of the ECG signal from the image after performing the decoding stage. Both the cubic spline and the median edge detection (MED) predictor have been adopted to conduct the interpolation process for estimating image pixels. Intensive experiments have been conducted to evaluate the performance of the multimodal scheme using objective distortion criteria. Results show clear superiority of the proposed scheme over the conventional separate compression approach involving two codecs: JPEG-2000 for images and ECG SPIHT-1D as well as other competing multimodal compression schemes in terms of both PRD and SNR at the signal decompression stage while maintaining good image quality and exhibiting a reduced computational complexity. Improvements in terms of average PRD and SNR values are as significant as 0.7 and 6 dB at low bit rates and 0.06 and 2 dB at higher bit rates on a number of test ECG signals and medical images.

Fast Hyperspectral Image Encoder Based on Supervised Multimodal Scheme

Many compression methods, lossy or lossless, were developed for 3D hyperspectral images, and various standards have emerged and applied to these amounts of data in order to achieve the best rate-distortion performance. However, high-dimensional data volume of hyperspectal images is problematic for compression and decompression time. Nowadays, fast compression and especially fast decompression algorithms are of primary importance in image data applications. In this case, we present a lossy hyperspectral image compression based on supervised multimodal scheme in order to improve the compression results. The supervised multimodal method is used to reduce the amount of data before their compression with the 3D-SPIHT encoder based on 3D wavelet transform. The performance of the Supervised Multimodal Compression (SMC-3D-SPIHT encoder) has been evaluated on AVIRIS hyperspectral images. Experimental results indicate that the proposed algorithm provides very promising performance at low bit-rates while reducing the encoding/decoding time.

Lossless Compression of Binary Trees with Correlated Vertex Names.

Compression schemes for advanced data structures have become a central modern challenge. Information theory has traditionally dealt with conventional data such as text, images, or video. In contrast, most data available today is multitype and context-dependent. To meet this challenge, we have recently initiated a systematic study of advanced data structures such as unlabeled graphs [8]. In this paper, we continue this program by considering trees with statistically correlated vertex names. Trees come in many forms, but here we deal with binary plane trees (where order of subtrees matters) and their non-plane version (where order of subtrees doesn't matter). Furthermore, we assume that each name is generated by a known memoryless source (horizontal independence), but a symbol of a vertex name depends in a Markovian sense on the corresponding symbol of the parent vertex name (vertical Markovian dependency). Such a model is closely connected to models of phylogenetic trees. While in general the problem of multimodal compression and associated analysis can be extremely complicated, we find that in this natural setting, both the entropy analysis and optimal compression are analytically tractable. We evaluate the entropy for both types of trees. For the plane case, with or without vertex names, we find that a simple two-stage compression scheme is both efficient and optimal. We then present efficient and optimal compression algorithms for the more complicated non-plane case.

An improved multimodal signal-image compression scheme with application to natural images and biomedical data

In this paper, a new multimodal compression scheme is proposed with the aim of compressing jointly an image and a signal via a single codec. The key idea behind our approach is to insert a wavelet-decomposed signal into a decomposed image and then consider the mixture data as an image for compression with the Set Partitioning In Hierarchical Trees (SPIHT) encoder. The insertion stage is performed in detail wavelet sub-bands using a spiral insertion function. The evaluation process is assessed on both natural and medical images according to an objective and subjective comparison criteria. Moreover, four multimodal compression schemes are provided for the sake of fair assessment. Finally, experimental results demonstrate the effectiveness of the proposed approach to achieve significant gains in terms of Percentage of Root Mean Square Difference (PRD) and Peak Signal to Noise Ratio (PSNR) for both reconstructed signal and image.

On Staying Grounded and Avoiding Quixotic Dead Ends.

The 15 articles in this special issue on The Representation of Concepts illustrate the rich variety of theoretical positions and supporting research that characterize the area. Although much agreement exists among contributors, much disagreement exists as well, especially about the roles of grounding and abstraction in conceptual processing. I first review theoretical approaches raised in these articles that I believe are Quixotic dead ends, namely, approaches that are principled and inspired but likely to fail. In the process, I review various theories of amodal symbols, their distortions of grounded theories, and fallacies in the evidence used to support them. Incorporating further contributions across articles, I then sketch a theoretical approach that I believe is likely to be successful, which includes grounding, abstraction, flexibility, explaining classic conceptual phenomena, and making contact with real-world situations. This account further proposes that (1) a key element of grounding is neural reuse, (2) abstraction takes the forms of multimodal compression, distilled abstraction, and distributed linguistic representation (but not amodal symbols), and (3) flexible context-dependent representations are a hallmark of conceptual processing.

Multimodal compression applied to biomedical data

In this paper, we introduce a novel approach to compress jointly a medical image and a multichannel biosignals (e.g. ECG, EEG). This technique is based on the idea of Multimodal Compression (MC) which requires only one codec instead of multiple codecs. Objectively, biosignal samples are merged in the spatial domain of the image using a specific mixing function. Afterwards, the whole mixture is compressed using JPEG 2000. The spatial mixing function inserts samples in low-frequency regions, defined using a set of operations, including down-sampling, interpolation, and quad-tree decomposition. The decoding is achieved by inverting the process using a separation function. Results show that this technique allows better performances in terms of Compression Ratio (CR) compared to approaches which encode separately modalities. The reconstruction quality is evaluated on a set of test data using the PSNR (Peak Signal Noise Ratio) and the PRD (Percent Root Mean Square Difference), respectively for the image and biosignals.

Fast Encoding-Decoding of 3D Hyperspectral Images Using a Non-Supervised Multimodal Compression Scheme

We introduce in this paper an extension of the Multimodal Compression technique (MC) for the purpose of coding hyperspectral image sequences. The main idea requires few steps, namely: (1) reducing the size of the sequence by inserting smooth images containing less information into the remaining images of the same sequence, (2) then coding the new compacted sequence using 3D-SPIHT algorithm. In this new scheme, called MC-3D-SPIHT, the insertion is achieved only in the contour of each image, according to a non-supervised way, so that one can preserve the Region of Interest (ROI) quality. For this purpose, a mixing function is employed. After the decoding process, inserted images are extracted by a separation function and the original sequence is reconstructed. By considering data from AVIRIS database, we will show how one decrease significantly the computing time for both coding and decoding.

Multimedia Information Compression Technologies

<em>Introduction</em> We are drowning in data. What kinds of data? - Text. Images. Sound. Numeric. Genome data. Text: Every day vast amounts of textual data are generated. This ranges from private corporate data, personal information, public and private government documents and so on. Much of this data needs to be accessed by many users for many tasks. For example, a corporate call centre needs fast access to documents at a semi-concept level to answer user requests. Another example: large litigations can involve 2 million documents, 200,000 of which are relevant, much fewer significant, and a handful pivotal. Techniques are desperately needer to automate the first few steps of this winnowing. Images: There are video cameras everywhere, trying to protect our safety in car parks, public places, even some lifts. There are huge and ever growing still and video archives of all aspects of our modern world. Access and indexing this data is a huge research enterprise. Much indexing is done manually. Sound: Often in concert with video in multi-media recordings. But what did the Prime Minister say on the 1st of November about the Republic? Did he sound like he meant it? These are currently not easily answered queries except if carried out by an expert human investigator. These kind of queries will need to be commonplace to access sound data in humanly meaningful ways. Numeric: Our industries generate vast amounts of valuable numeric data. In the petroleum industry geologic knowledge must be integrated with data from wells: laboratory core analysis data and on-site well logs, with seismic data generated from controlled explosions and dispersed recording devices. Then there is GIS data collected from satellites and so on. In the service industry, the stock exchange generates large amounts of hard to analyse data vital to the wellbeing of Australian companies. Genome data: The human genome project is almost complete. Researchers are finding genes by a mix of laboratory work and computerised database searches (e.g. as reported in the Weekend Australian 30 October). This is just the first step, the next will be sequencing of a number of individuals, and of course there are currently over 100 whole genome sequencing projects on other species. Fast genome sequencing is just around the corner. We will soon be drowning in this kind of data also. Multimedia data: Includes all of audio, text, graphics, images, video, animation, music. More data! <em>What Is The Real Problem?</em> Manual extraction of information from any large corpus is time con-suming and expensive, requiring specialised experience in the material. Even worse, beyond a certain point it is incredibly boring, and hence error prone. Human intelligence is best suited to dealing with information, as distinct to data! <em>A Solution</em> The development of automated systems for information extraction, and for the synthesis of the extracted information into humanly useful information resources. To avoid drowning in the ever increasing flow of multi-modal electronic information available, automated tools are required to reduce the cognitive load on users. <em>STEPS TOWARDS A SOLUTION</em> The key step towards a solution is the notion of information compression, being the compression of data to yield an information rich(er) resource. This is distinct from data compression which is merely the efficient storage of data. Further, the information compression must work on multi-model complex data, exemplified by multimedia data. Some of the techniques for doing this kind of information compression exist in a scattered way in areas such as fuzzy systems, and image analysis. We have identified a nascent field, which we can coalesce in an intensive short workshop. The first Australia-Japan Joint Workshop on Applications of Soft and Intelligent Computing to Multimodal and Multimedia Information Compression Technologies was held at Murdoch University in Perth, Western Australia from the 29 March to 5 April 2000. This special issue contains selected papers from the workshop.