Arun Ramaswamy

A mixed transform approach for efficient compression of medical images

A novel technique is presented to compress medical data employing two or more mutually nonorthogonal transforms. Both lossy and lossless compression implementations are considered. The signal is first resolved into subsignals such that each subsignal is compactly represented in a particular transform domain. An efficient lossy representation of the signal is achieved by superimposing the dominant coefficients corresponding to each subsignal. The residual error, which is the difference between the original signal and the reconstructed signal is properly formulated. Adaptive algorithms in conjunction with an optimization strategy are developed to minimize this error. Both two-dimensional (2-D) and three-dimensional (3-D) approaches for the technique are developed. It is shown that for a given number of retained coefficients, the discrete cosine transform (DCT)-Walsh mixed transform representation yields a more compact representation than using DCT or Walsh alone. This lossy technique is further extended for the lossless case. The coefficients are quantized and the signal is reconstructed. The resulting reconstructed signal samples are rounded to the nearest integer and the modified residual error is computed. This error is transmitted employing a lossless technique such as the Huffman coding. It is shown that for a given number of retained coefficients, the mixed transforms again produces the smaller rms-modified residual error. The first-order entropy of the error is also smaller for the mixed-transforms technique than for the DCT, thus resulting in smaller length Huffman codes.

Region based variable quantization for JPEG image compression

Introduces concepts that optimize image compression ratio by utilizing the information about a signals properties and their uses. This additional information about the image is used to achieve further gains in image compression. The techniques developed in this work are on the ubiquitous JPEG still image compression standard [IS094] for compression of continuous tone grayscale and color images. This paper is based on a region based variable quantization JPEG software codec that was developed tested and compared with other image compression techniques. The application, named JPEGTool, has a graphical user interface (GUI) and runs under Microsoft Windows (R) 95. This paper discusses briefly the standard JPEG implementation and software extensions to the standard. region selection techniques and algorithms that complement variable quantization techniques are presented in addition to a brief discussion on the theory and implementation of variable quantization schemes. The paper includes a presentation of generalized criteria for image compression performance and specific results obtained with JPEGTool.

Residual error formulation and adaptive minimization for representing nonstationary signals using mixed transforms

A technique is proposed for signal representation using superimposed partial sets of different transforms which are, in general, nonorthogonal to each other. The method is developed to maximize the signal-to-noise ratio (SNR) of the reconstructed signal for a given total number of transform coefficients. First, the residual error, which is the difference between the original signal and the reconstructed signal, is properly formulated. Then, two gradient techniques, in conjunction with an optimization strategy, are developed to minimize the residual error. Sample results using this approach for representing synthetic signals and speech signals employing mixed Fourier/Walsh and Fourier/Haar transforms are given to illustrate the efficiency and accuracy of the proposed method. >

Application of multitransforms for lossy image representation

The multitransform technique is applied for lossy image representation. The image is first divided into smaller non-overlapping subimages. Each subimage is resolved into two dimensional (2-D) subsignals, which are then compactly represented in a particular transform domain. This leads to an efficient representation of the subimage by superimposing the dominant coefficients corresponding to each subsignal. An adaptive algorithm in conjunction with an optimization strategy are developed to minimize the difference between the original and the reconstructed subimage. Extensive simulations verify that, for a given number of retained coefficients, the combination of the DCT and Haar transforms yields a more compact representation than using DCT or Haar alone. >

An efficient representation of nonstationary signals using mixed-transforms with applications to speech

A useful technique is presented to efficiently represent nonstationary signals by combining the wavelet and mixed-transforms. First, the signal is split into subbands using the discrete wavelet transform. Specifically, the subband splitting is performed by employing the scaling and the wavelet functions to low-pass and high-pass filter the signal. Next, each of these subbands is represented using superimposed partial sets of basis functions of different transforms, which are in general, mutually nonorthogonal. This is termed the mixed-transforms representation. The residual error, which is the difference between the original subband and the reconstructed subband is properly formulated. Adaptive algorithms are developed to minimize this error and to maximize the Signal to Noise ratio (SNR) of the reconstructed subband for a fixed number of transform components. An optimization strategy is also proposed to select the dominant components from the various domains for adaptation. These efficiently represented subbands are finally combined to reconstruct the signal. Sample results for representing voiced and unvoiced speech signals are given to illustrate the accuracy of the proposed method. A specific class of scaling and wavelet functions are employed in conjunction with the mixed Fourier/Haar transforms. It is verified that for a given number of coefficients, the proposed technique yields a significantly higher SNR of the reconstructed signal than using mixed-transforms alone or the wavelet transform followed by a single transform. Performance comparison is also presented for two different orders of scaling functions belonging to the same family. >

A standard target decoder model for MPEG-4 FlexMux streams

MPEG-4, like other MPEG standards is generic and universal in the sense that it specifies compressed bitstream syntax. This, in effect, unambiguously defines the decompression process and the decoder architecture. Further, to ensure interoperability between encoders and decoders, it is essential that a standard target decoder model (STD) be well defined. Since STD is a hypothetical model in the encoder architecture, multiplexors can create compliant streams without having to base their design around any particular decoder. At the same time, compliant decoders should have the minimum buffers as mandated by the STD and hence should be able to decode compliant streams. This paper proposes an STD model for MPEG-4 FlexMux framework.

COMPRESSION OF IMAGES REPRESENTED USING MULTI-TRANSFORMS

Constrained image representation employing multi-transforms has been recently developed. First, the image is divided into smaller non-overlapping subimages. Each subimage is resolved appropriately into 2-D subsignals, each of which is compactly represented in a specific transform domain. The subimage is efficiently represented by superimposing the dominant components corresponding to the subsignals. The residual error, which is the difference between the original subimage and the reconstructed subimage is minimized by adaptive algorithms. An optimization strategy selects the dominant coefficients from the various domains for adaptation. An efficient coding technique is presented to code the multi-transform coefficients. An image representation example is presented employing the DCT-Haar combination. Objective evaluations are made where it is shown that images represented using the multi-transform technique are more accurate than using the DCT for the same number of retained transform coefficients. Test su...

An efficient coding technique for multi-transform image representation

Constrained image representation employing multi-transforms has been recently developed. An efficient coding technique is presented in this paper to code the multi-transform coefficients. An image representation example is presented employing the DCT-Haar combination. The bandlimited nature of the DCT and the Haar spectra allows the scheme to be similar to the JPEG baseline standard. Test subimages with high amount of detail represented using the proposed technique show a SNR improvement of about 3 to 4 dB over using DCT alone. Finally, images, coded at bit rates of 0.44 bits/pixel and 1.23 bits/pixel employing the proposed technique verify the good quality of reconstruction.<<ETX>>

Multitransform/multidimensional signal representation

A useful technique is presented to compactly represent multidimensional signals employing two or more transforms. This is performed by superimposing partial sets of basis functions of different transforms, which are in general, mutually non-orthogonal. Basically, the multidimensional signal is split into subsignals, each of which is represented by the dominant components of a transform, whose basis functions closely approximate the subsignal. The residual error, which is the difference between the original image and the reconstructed image is properly formulated. Adaptive algorithms are developed to minimize this error and to maximize the Signal to Noise ratio (SNR) of the reconstructed image for a fixed number of transform components. An optimization strategy is also proposed to select the dominant components from the various domains for adaptation. A simulation is presented to represent a synthetic image consisting of two sinusoids and two square waves. It is verified that it is possible to split the image spectrally into a narrowband and a broadband part, thus allowing the DCT to represent the sinusoids and the Walsh Hadamard transform to represent the square waves. This leads to excellent data compression.<<ETX>>

Constrained image representation using multi-transforms

A technique is presented to represent images employing two or more mutually non-orthogonal transforms. First, the image is divided into smaller non-overlapping subimages. Then, each subimage is resolved into two dimensional (2-D) subsignals such that each of the subsignals is compactly represented in a particular transform domain. This leads to an efficient representation of the subimage by superimposing the dominant coefficients corresponding to each subsignal. The residual error, which is the difference between the original subimage and the reconstructed subimage is properly formulated. An adaptive algorithm in conjunction with a strategy is developed to minimize this error for a fixed number of transform coefficients. The feasibility of the method is established through a synthetic image example. It is also shown that the DCT-Haar combination offers a percentage reduction of up to 15.0 to 17.0% in the Root Mean Square Error (RMSE) for the test images as compared to the DCT alone.<<ETX>>