A New Two Successive Process Image Compression Technique Using Subband Coding and JPEG Discrete Cosine Transform Coding
✍ Scribed by C.P. Liu
- Publisher
- Elsevier Science
- Year
- 1997
- Tongue
- English
- Weight
- 646 KB
- Volume
- 59
- Category
- Article
- ISSN
- 1077-3169
No coin nor oath required. For personal study only.
✦ Synopsis
the resulting subimages are then down-sampled and en-This paper proposes a new image compression technique coded for transmission. At the receiver, the individual subbased on successive application of a 2-D single-sideband analyimages are decoded, up-sampled, and filtered to allow the sis/synthesis system and the Joint Photographic Experts Group recovery of the original image. Note that in the synthesizer, (JPEG) discrete cosine transform (DCT) lossy transform coder. the corresponding signal will be denoted by the ''Ј'' nota-A 2-D separable single-sideband (SSB) analysis/synthesis systion.
which is developed in terms of a 2-D separable weighted R Ϫ 1 denotes the input channel signals to the synthesizer overlapped-add method of analysis/synthesis and which allows and xЈ(m, n) denotes the output signal. Since the coding overlap between adjacent spatial domain windows, is used first distortion generated by the encoding (or quantization) of to reduce the image size in the spatial domain. The JPEG the decimated signals is introduced during transmission, discrete cosine transform is then used to reduce the image size in the frequency domain. These two successive compression the input signal to the synthesis channel X Ј k,t (m, n) is reprocesses combine to form a powerful image compressor. The lated to, but not always equal to, the output of the analysis overall compression of images in this technique can reach up channel signal X Ј k,t (m, n). However, reconstruction of the to about 97 percent of their original size without much of the signal x(m, n) is possible only if the coding distortion has image quality being lost. © 1997 Academic Press been eliminated.
The proposed analysis/synthesis system is based on a block transformation. The analysis process involves win-
I. INTRODUCTION
dowing the signal and transforming the windowed signal using a frequency domain transform that includes the dis-Image compression is the efficient coding of digital imcrete cosine transform (DCT), the discrete sine transform ages to reduce the number of bits required in storage and (DST), and the discrete Fourier transform (DFT). The transmission while maintaining acceptable fidelity or image inverse transforms are utilized during the synthesis process, quality. In this paper, we present an efficient implementaand the resulting spatial domain signal sequence is tion of two-dimensional SSB analysis/synthesis system multiplied by a synthesis window. The output response which is capable of transmitting good quality images at high compression rates [1-3]. The proposed system consists of is simply overlapped and added to the accumulator shift two distinct subsystems: (1) an analysis/synthesis filter register of previously accumulated signal segments [1-3]. which splits the signal (or image) into narrow bands, and Again, a straightforward expression for the requirements (2) a coder which encodes each subband signal for the of the analysis/synthesis system can be designed so that purpose of transmission. The first part subsystem is based the signal x(m, n) is properly reconstructed. The signal on a 2-D separable single-sideband analysis/synthesis sysx(m, n) can be reconstructed if the adjacent pairs of 2-D tem in a spatial domain which is an extension from the analysis/synthesis window responses overlap and add in 1-D case [1]. Two-dimensional (2-D) (analysis/synthesis) the spatial domain so that the result is flat. We will see in filter banks have found many applications in video coma later section that any spatial domain aliasing introduced pression and image coding. A basic framework for a 2-D in the analysis will be cancelled during the synthesis. analysis/synthesis filter system is shown in Fig. 1. A 2-D
In the design of analysis/synthesis systems, it is desired signal, such as an image, is first split into subimages, and that the overall sampling rate at the output of the analysis bank be equal to that of the input signal. There are three sampling rates that can be applied to the system: (1) criti-