This paper presents a novel topology for the even harmonic mixer (EHM). The proposed mixer employs current reuse and double frequency circuits in the RF input stage and LO stage, respectively, to improve its linearity and isolation. In addition, the proposed topology has the advantage of the low power consumption. In order to demonstrate the benefits of the proposed mixer, theoretical analyses of conversion gain and linearity have been described in details. The measured results reveal that the proposed mixer possesses single-end conversion gain of 8 dB and third-order input intercept point (IIP 3) of –3.8 dBm, respectively, under the supply voltage of 1.8 V and LO power of 4 dBm. The power consumption of the proposed mixer is about 1.4 mW at 900 MHz.

In the proposed model, dynamic inhibitory weights are used to speed up the convergence rate, and a new convergence rule is applied to find all maxima. The hardware implementation of the proposed model is presented in the study, and simulation results indicate that the proposed model converges much faster than the other networks.

Facial model coding is an integral part in MPEG-4 related applications. The generation of the facial model usually requires stereoscopic view of the face in the pre-processing stage. Although facial model can be successfully estimated from two stereo facial images, the occlusion effect and imprecise location of the feature point prohibit obtaining an accurate facial model. In this paper, several facial model estimation (FME) algorithms are proposed in order to find the precise facial model from a stereo or mono image sequence. Since a sequence of images is used to find the facial model, the problem of occlusion effects is less serious. An accurate facial model (within 7.21% error) can still be obtained by our schemes, even without the prior information on the three-dimensional position of the head with respect to the camera and the rotation axis/angle of the head's movement. This is the largest error of all FME algorithms presented in this paper when the subject does not wear eyeglasses. In addition, our schemes do not require precise camera parameters and avoid tedious camera calibration, thereby, simplifying the facial model extraction.

Image database systems must effectively and efficiently handle and retrieve images from a large collection of images. A serious problem faced by these systems is the requirement to deal with the nonstationary database. In an image database system, image features are typically organized into an indexing structure, and updating the indexing structure involves many computations. In this paper, this difficult problem is converted into a constrained optimization problem, and the iteration-free clustering (IFC) algorithm based on the Lagrangian function, is presented for adapting the existing indexing structure for a nonstationary database. Experimental results concerning recall and precision indicate that the proposed method provides a binary tree that is almost optimal. Simulation results further demonstrate that the proposed algorithm can maintain 94% precision in seven-dimensional feature space, even when the number of new-coming images is one-half the number of images in the original database. Finally, our IFC algorithm outperforms other methods usually applied to image databases.

Image coding based on a linear transform has been extensively studied. On the other hand, the only nonlinear transform used for image coding, to our best knowledge, is the fractal coding. We proposed a rigorous coupled wave (RCW) nonlinear transform for image coding. Here, we model the image as a surface relief grating where the gray-scale value of the image pixel is directly converted to the pixel's depth of the grating. When a uniform light wave is shined through the grating, the amplitude of the backward- and forward-diffracted waves are quantized and recorded, while the phase of the backward- and forward-diffracted waves can be calculated based on some a priori boundary conditions. Then these quantized diffraction waves can be used to reconstruct the depth of the grating (or image signal) based on Maxwell's equations. Scalar quantizations (SQ) combined with variable length coding are tested. Respectively, the PSNR of the SQ coded RCW transformed image at a low bit rate (0.1 bit/pixel) is 30.83 dB, while the state-of-the-art JPEG-2000 only achieves 28.61 dB.

Arbitrarily shaped coding is an important tool to achieve MPEG-4, which is a object-based coding standard. We proposed an efficient shaped coding method which enhances the coding efficiency of conventional padding techniques in the object-based video coding where the shape information is provided. The boundary block-searching algorithm (BBS) is applied to the boundary blocks of a VOP (video object plane) which consists of both background and object pixels. Actually, boundary blocks have strong correlation even though they are located apart. For an input boundary block, we search the closest block (only object pixels) from the previously encoded data using the BBS algorithm. After searching, the boundary block is represented as a position vector. This process reduces the number of boundary blocks to be DCT and makes the bit rate small. Experimentation has been done on various video sequences under different test conditions and verifies significant coding efficiency improvement.

A 3-D facial model can be accurately obtained from: (1) stereo facial images or (2) front- and side-view facial images. The stereo images can be obtained by two different cameras at the same time. On the other hand, capturing front- and side-view facial images requires precise location of the camera. Either may be unavailable in many practical applications. We modify several existing techniques to automatically locate the feature point position from the front-view facial image. Then several schemes (such as minimum mean square error, minimum mean absolute error, and maximum a posteriori) are used to estimate the "depth" information of the 3-D facial model parameters from the single front-view facial image according to the anthropometric and a priori information. With this scheme, a 3-D facial model can be estimated from single front-view facial image. According to the simulation results, the estimated facial model matches the exact 3-D facial model, and the synthesized and original side-view facial images appear similar.

Modified network architecture for sorting data of any size is presented in this paper. The proposed VLSI archi- tecture is directly modified from the Batcher's odd-even merge-sort network. The major advantage of the proposed architecture is its modular approach, simple and regular interconnection and easy implementation by VLSI technol- ogy.

There is a need to downscale the video (from image size to ) prior to transmission due to bandwidth limitation on the cannel. The conventional method for downscaling the video sequence is to decompress the video bitstream and then reencode it again at video server. This process is computationally intensive because of the motion estimation required during the reencoding process. In this paper, we proposed a polynomial motion-vector resampling (PMVR) algorithm to compute motion vectors for the downscaled video sequence in compressed domain. The proposed approach leads to significant computational saving compared with the conventional spatial domain approach. Simulation results suggest that the proposed method is effective for video downscaling.

This paper proposes a polynomial search (PS) algorithm and architecture to solve the motion estimation problem in video coding. Simulation results show that the proposed method is not only flexible, but also requires fewer computations to achieve the same mean absolute error results (for QCIF and sub-QCIF video) compared with the existing fast-search algorithms. Finally, a VLSI architecture is also developed to efficiently implement the PS algorithm.

The domain pool design is one of the dominant issues which affect the coding performance of fractal image compression. In this paper, we employ the LBG algorithm and propose a block averaging method to design the efficient domain pools based on a proposed iteration-free fractal image codec. The redundancies between the generated domain blocks are reduced by the proposed methods. Therefore, we can obtain the domain pools that are more efficient than those in the conventional fractal coding schemes and thus the coding performance is improved. On the other hand, the iteration process in the conventional fractal coding scheme not only requires a large size of memory and a high computation complexity but also prolongs the decoding process. The proposed iteration-free fractal codec can overcome the problems above. In computer simulation, both the LBG-based and block-averaging methods for the domain pool design in the proposed iteration free scheme achieve excellent performances. For example, based on the proposed block-averaging method, the decoded Lena image has at least a 0.5 dB higher PSNR (under the same bit rate) and an eight-time faster decoding speed than the conventional fractal coding schemes that require iterations.

In an object-based analysis-synthesis coder (OBASC) for teleconference, an object can be represented by a model of triangle mesh wireframe. Because the video contents usually are the head-and-shoulders or heads only, the generation of a facial model and its accuracy thus are important issues. A three-dimensional facial model can be generated using a stereo image pair and the knowledge of the disparity. But a facial model reconstructed from a stereo image pair only may be inaccurate because of the occlusion of some features. A new facial model refinement algorithm is proposed in this paper. Here we first generate a rough facial model using the first stereo image pair in a stereo image sequence, and then fine tune this model according to the further information gathered from the following image sequence. Our results converge to the exact facial model and thus provide a better model for the OBASC.

Noise reduction of VQ encoded images is achieved through the proposed anti-gray coding (AGC) and noise detection and correction scheme. In AGC, binary indices are assigned to the codevector in such a way that the 1-b neighbors of a code vector are as far apart as possible. To detect the channel errors, we first classify an image into uniform and edge regions. Then we propose a mask to detect the channel errors based on the image classification (uniform or edge region) and the characteristics of AGC. We also mathematically derive a criterion for error detection based on the image classification. Once error indices are detected, the recovered indices can be easily chosen from a "candidate set" by minimizing the gray-level transition across the block boundaries in a VQ encoded image. Simulation results show that the proposed technique provides detection results with smaller than 0.1% probability of error and more than 86.3% probability of detection at a random bit error rate of 0.1%, while the undetected errors are invisible. In addition, the proposed detection and correction techniques improve the image quality (compared with that encoded by AGC) by 3.9 dB.

We propose an approach to efficiently compress and transmit multiresolution 3D facial models for multimedia and very low bit rate applications. A personal facial model is obtained by a 3D laser digitizer, and further re-quantized at several resolutions according to different scope of applications, such as animation, video game, and video conference. By deforming 2D templates to match and re-quantize a 3D digitized facial model, we obtain its compressed model. In the present study, we create hierarchical 2D facial wireframe templates at five resolutions. In the deformation procedure, 2D templates are adapted according to facial feature points and the proposed piecewise chainlet affine transformation (PCAT) method. The 3D digitized model after requantization are reduced significantly without perceptual loss. Moreover, the proposed multiresolution facial models possessed a hierarchical data structure and are apt to be progressively transmitted and displayed across the Internet.

Based on the previously developed sorting networks, a new VLSI architecture suitable for two-dimensional (2-D) rank-order filtering is proposed. The major advantage of the proposed architecture is their fast response time, modular architecture, and simple and regular interconnection. Generally speaking, the throughput of the proposed architecture is (N-1) times faster than using a one-dimensional rank-order filter for 2-D N/spl times/N data. The concept of block processing is also incorporated into the design to reduce the time-area complexity of the proposed architecture. Roughly speaking, the complexity is reduced to 2/3 and 1/2 compared with a rank-order and median filter without using a block processing architecture, respectively. A 3/spl times/3 median filter with block processing architecture is implemented through a 0.8 /spl mu/m single-poly double-metal CMOS process. The results are correct with a clock rate up to 125 MHz.

An adaptive technique for scanning rate conversion and interpolation is proposed. This technique performs better than the edge based line average algorithm, especially for an image with more horizontal edges. Moreover, it is easy to implement and a simple VLSI architecture is proposed in this paper. Computer simulation shows that a 37.0 dB image can be obtained via our proposed technique, while edge based line average algorithm only achieve 35.2 dB.

In anti-Gray codes, binary indexes are assigned to code vectors in a way that maximizes the average quantization distortion introduced in the reproduced source vectors when a transmitted index is corrupted by channel noise. In this paper, anti-Gray coding is used for VQ encoded images. Since the average distortion at the receiver is maximized by the anti-Gray coding technique, errors can be easily detected. To detect channel errors, we first classify an image into uniform and edge regions. Then we propose a mask to detect channel errors based on the image classification (uniform and edge regions) and anti-Gray code characteristics. We also mathematically derive criteria for error detection based on image classification. Once error indexes are detected, the recovered indexes can be easily chosen from a candidate set by minimizing the gray-level transition across the block boundaries in a VQ encoded image. Simulation results show that the proposed technique provides the detection results with less than 0.1% probability of error and more than 86.3% probability of detection at random bit error rate 0.1%. For those undetected errors, they appear invisible. In addition, the proposed approach provides a better PSNR performance than other similar techniques and pseudo-Gray codes for images under noisy channels.

The order statistic filter is an important nonlinear filter for many image processing applications. We propose several simplified algorithms for the VLSI implementation of an order statistic filter. Specifically, we propose: (1) a simplified algorithm to reduce the chip area of an order statistic filter; (2) a simplified algorithm using fewer components compared with the Murthy and Swamy's (see IEEE Transactions on Signal Processing, vol.40, no.5, p.1241-52, 1992) results to integrate two r-bit order statistic filters into an (r+1)-bit order statistic filter; and (3) a scheme for 2-D order statistic filter implementation. In this scheme a large ROM is replaced by very few logic components. In summary, our schemes reduce the chip area of an order statistic filter because much fewer components are used compared with the Murthy and Swamy's results.

An image compression scheme based on the finite-state fractal block coding (FS-FBC) is proposed. Since the FS algorithm has been successfully employed by the vector quantization (VQ) techniques and there are many similarities between VQ and FBC scheme, we employ the FS algorithm on FBC technique to determine the fractal codes. The next-state functions for the domain pool size, contrast scaling factor, isometry, and mean are designed. In computer simulation, we first independently investigate the FS effect of each parameter and then combine all methods to test the overall effect. We obtain a reconstructed Lena image with 33.9 dB PSNR at 0.42 bit/pixel, while the result of original FBC scheme is 31.4 dB at 0.6 bit/pixel. The simulation results show that our FS methods can be successfully employed on the FBC technique to reduce the bit rate and the image quality is still preserved.

Based on the sorting networks, a new VLSI architecture suitable for 2-D rank order filtering is proposed. The major advantage of the proposed architecture is the fast response time and modular architecture. Generally speaking, the throughput of the proposed architecture is (N-1) times faster than using a 1-D rank order filter for 2D N/spl times/N data. The concept of block processing is also incorporated into the design to reduce the time-area complexity of the proposed architecture. Roughly speaking, the complexity is reduced to 2/3 and 1/2 compared with a rank order and median filter without using block processing architecture, respectively. A 3/spl times/3 median filter with block processing architecture is implemented through a 0.8 /spl mu/m single-poly double metal CMOS process. The simulation results are correct with a clock rate up to 91 MHz.

Fractal image coding is an effective method to eliminate the image redundancy through piecewise self transformability called fractal codes on a block-wise sequence. The fractal code is a set of contractive affine transforms defined as a sequence of parent and child block transformations. To improve the performance when a parent range block experiences large error, we usually partition the parent range block into four children (smaller) square blocks. In this paper, a shape-adaptive fractal coding (SAFC) based on the edge information is proposed. In SAFC, each parent range block is first partitioned into nonsquare (rectangle or triangle) children blocks which are to be fractal encoded. If the coded image quality does not improve, then the traditional square partition is executed. In SAFC, no additional computations are required to obtain the edge orientation of a range block. Instead, the edge orientation (of a range block) is obtained during the isometries calculation between a range block and all its possible domain blocks. According to our simulation, SAFC achieve 33.9 dB PSNR at 0.49 bit/pixel (bpp), while the conventional fractal coding achieves 34.0 dB at 0.55 bpp. In addition, we mathematically show that the proposed detector does find the correct edge orientation of blocks under certain conditions.

Side-match vector quantization is a finite-state technique for image coding. This research shows that the side-match vector quantization is an error-propagating code and it is similar to a catastrophic convolutional code. Here, we propose a Viterbi-based algorithm to solve this problem. Various noise detection algorithms are integrated into the Viterbi algorithm (to yield the Viterbi-based algorithm) for a much better performance. According to the simulation of a binary symmetric channel with random bit-error rate (BER) 0.1%-0.01%, the Viterbi-based algorithm provides 2.8-9.6 dB and 2.3-8.4 dB gain compared with the conventional side-match vector quantization decoder and the Viterbi decoder, respectively. In addition, the proposed algorithm requires much fewer computations than the Viterbi algorithm.

An adaptive technique for scanning rate conversion and interpolation is proposed. This technique performs better than the edge-based line average algorithm, especially for an image with more horizontal edges. Moreover, it is easy to implement and a simple VLSI architecture is proposed. Computer simulation shows that a 37.0 dB image can be obtained via our proposed technique, while edge-based line average algorithm only achieves 35.2 dB.

Blocking effects are a major degradation in block based image coding techniques. In this paper, we propose an adaptive postprocessor for the removal of blocking effects. The adaptation is achieved by changing filter coefficients according to the local characteristics of images and the blocking effects. Here, a space-variant lowpass filter is used to smooth the pixels at block boundaries where the blocking effects are highly visible. In addition, the pixels that are closed to an edge are adaptively regional or directional lowpass filtered if the blocking effects are obvious. As for the pixels at edges, they are left untouched by the postprocessor. Therefore, blocking effects can be removed without blurring the edge sharpness. Performance improvements compared with other techniques are obtained according to simulation results. The proposed postprocessor has 0-1 dB gain (depending on the images) compared with the loop filter of H.261, the linear filter, and the nonlinear space-variant postprocessor.>

The quantization efficiency of independent and uncorrelated random variables is evaluated. The uncorrelated and independent random variables are generated by Karhunen-Loeve (K-L) transform of the natural scene image and the encrypted image, respectively. As for the encrypted image samples, they are defined as the weighted sum of the natural scene image samples. After transform, optimal scalar and vector quantization is then performed on these transform coefficients. Simulation results show that the performance of scalar quantization of the independent random variables increases compared with that for the uncorrelated ones. Since the encrypted image can be easily generated by fast Fourier transform, the desirability of using vector quantization decreases. Because vector quantization is much more difficult to implement compared with scalar quantization, the performance improvement of scalar quantization by encryption is a feasible solution.>

Two-dimensional quasi-maximal-area arrays (quasi-m-arrays) are proposed. These quasi-m-arrays are so named because their cyclic correlation properties are very close to those of the m-arrays. These arrays are generated by modulo-2 addition of maximal length sequences. The cyclic autocorrelation of any quasi-m-array is close to the delta-function. In addition, the cyclic cross-correlation is small compared with the cyclic autocorrelation peak. Two-dimensional Gold code arrays generated by the same construction method are also studied. Correlation properties of these Gold code arrays are similar to those of the quasi-m-arrays. The Gold code arrays provide families of arrays in which any two arrays have the bound on their cyclic cross-correlation. These arrays are also shown to have the quasi-orthogonal property.>