Strip based coding for large images using wavelets

https://doi.org/10.1016/S0923-5965(02)00019-XGet rights and content

Abstract

Thanks to advances in sensor technology, today we have many applications (space-borne imaging, medical imaging, etc.) where images of large sizes are generated. Straightforward application of wavelet techniques for above images involves certain difficulties. Embedded coders such as EZW and SPIHT require that the wavelet transform of the full image be buffered for coding. Since the transform coefficients also require storing in high precision, buffering requirements for large images become prohibitively high. In this paper, we first devise a technique for embedded coding of large images using zero trees with reduced memory requirements. A ‘strip buffer’ capable of holding few lines of wavelet coefficients from all the subbands belonging to the same spatial location is employed. A pipeline architecure for a line implementation of above technique is then proposed. Further, an efficient algorithm to extract an encoded bitstream corresponding to a region of interest in the image has also been developed. Finally, the paper describes a strip based non-embedded coding which uses a single pass algorithm. This is to handle high-input data rates.

Introduction

Images acquired through remote sensing satellites are generally large in size. These images, acquired line-by-line by an optical sensor, are to be transmitted in compressed form on a constant bit rate channel to ground station for archival and analysis purpose.

Wavelet based compression techniques such as EZW [10] and SPIHT [9] have become benchmark techniques for state-of-art compression. Besides better compression and better image quality, they provide embedded bitstream, which can be truncated at any point and reconstruction carried out at varying quality. The price to be paid for these advantages is in the form of large memory requirements. For coding, the transform coefficients of full image need to be buffered in high precision, thus making memory requirements a bottleneck for hardware implementation. This problem is very serious especially in on board compression of satellite images, where memory is limited because of power constraints. While wavelet transform can be implemented by buffering only few lines [5], embedded coding is non-compatible with low memory environment. In this paper, we propose a strip based embedded coding (SBEC) for large images, with minimal buffering of wavelet coefficients. The image is processed line-by-line and the wavelet transform is performed over a set of scan lines. The wavelet coefficients are buffered in a strip buffer for embedded coding. The size of the strip is dependent on the image width and number of levels in wavelet decomposition, but independent of wavelet filter length and image height.

The current standard for still image compression, JPEG2000 [8] uses discrete wavelet transform for state-of-art compression. The standard employs embedded block coding [11] on wavelet subbands, whereas the proposed work uses zero-tree structure for embedded coding on strip images.

In some applications, user would like to retrieve only a specific region of an image (also known as region of interest (ROI) image) stored in compressed form. The user may be a remote client, requesting the ROI image. In this case, the general tendency is for server to decode the full image, extract the ROI image and encode, and transmit it to remote client. This is computationally expensive. An efficient algorithm to extract the encoded bit stream corresponding to ROI from the compressed bit stream is also proposed in this paper.

The paper is organized as follows. In Section 2, SBEC is described in detail. The strip buffer architecture for embedded coding and pipeline architecture for real-time implementation are presented along with performance details of the embedded coder. ROI image extraction from the compressed bitstreams and associated mathematical back ground are presented in Section 3. In Section 4, the non-embedded coding on the strip buffer is described. Section 5 concludes the paper.

Section snippets

SBEC of wavelet coefficients for large images

A separable wavelet transform can be implemented progressively using only 2L equivalent number of image lines, where L is the filter length, without requiring the full image to be buffered in memory. Embedded coding requires full wavelet image in the memory for coding. This is so because, even though the zero-tree structure used for embedded coding is formed with the wavelet coefficients belonging to same spatial location, all the zero trees in the wavelet image are scanned in each pass for

ROI image extraction from compressed domain for large images

When large images are archived in compressed form, extracting a specific subimage to view, also takes significant amount of time. The whole image need to be decoded and reconstructed. As the image size grows, the time taken also increases. If the request comes from a remote client, then the extracted image need to be encoded again for transmission. In this section, we describe an efficient algorithm to extract the encoded image bitstream corresponding to any ROI, directly from the compressed

Strip based non-embedded coding (SBNC)

Embedded coding techniques generally involve several passes. For real-time applications fast memories are required. Fast memories are generally expensive and power consuming. As we know the power consumption is a critical issue in satellite-based on-board applications. To overcome this problem, we have also proposed a strip based non-embedded coding SBNC. This algorithm can handle high-input data rates as each coefficient is read only once and processed.

As the 8×8 block size is standard in DCT,

Conclusion

In this work, we have presented a strip-based embedded coding for large images for real-time applications. This algorithm reduces the memory requirements significantly and retains all the advantages of embedded coding. The performance degradation is minimal, in view of reduced memory requirements. We have also developed a pipeline architecture for strip-based coding, for real-time applications. While the large images are compressed using the proposed SBEC, an efficient algorithm to extract the

Acknowledgements

The authors are grateful to Christos Chrysafis, Hewlett-Packard and Antonio Ortega, Integrated Media Systems Center for their executable code and test image provided.

References (13)

There are more references available in the full text version of this article.

Cited by (26)

  • Meta-scalable discriminate analytics for Big hyperspectral data and applications

    2021, Expert Systems with Applications
    Citation Excerpt :

    Note that the implementation also used divide-and-conquer mechanisms for the intermediate matrix multiplications (Thottethodiet al.,1998) for the strip-based processing. Some other example works where authors have proposed strip-based approaches to overcome implementation on limited memory platforms can be found in (ZainEldin et al., 2016; Chew et al., 2012; Bhattar et al., 2002). The dimensionality reduced data matrix after the GSLF-LDA was passed to two different classifiers (SVM and ensemble tree) to perform the classification tasks which returned recognition rates of 98.8% and 94.4% respectively.

  • A modified listless strip based SPIHT for wireless multimedia sensor networks

    2016, Computers and Electrical Engineering
    Citation Excerpt :

    The experimental results show that the proposed coding function requires 4504 Kb of memory during the processing as shown in Table 9. This means that, the proposed algorithm offers 71% memory reduction compared to SPIHT coder and 10% on average compared to strip-based SPIHT [21]. Finally, for the power consumption, Joulemeter is a software tool that estimates the power consumption (from CPU) of an application.

  • Scalable line-based wavelet image coding in wireless sensor networks

    2016, Journal of Visual Communication and Image Representation
  • Performance evaluation of the fractional wavelet filter: A low-memory image wavelet transform for multimedia sensor networks

    2011, Ad Hoc Networks
    Citation Excerpt :

    For completeness we briefly review this related research area, noting that to the best of our knowledge all techniques in this area have significantly higher memory requirements for the image wavelet transform than our approach. Building on [26], a strip based approached for low-memory image compression in proposed in [29]. In turn, building on [29], an approach for reducing the number of wavelet scales (transform levels) and an new transform tree structure to reduce the memory requirements for compression are proposed in [30,31].

  • Computationally efficient wavelet-based low memory image coder for WMSNs/IoT

    2023, Multidimensional Systems and Signal Processing
View all citing articles on Scopus
View full text