Jianwen Chen

FPGA-accelerated 3D reconstruction using compressive sensing

The radiation dose associated with computerized tomography (CT) is significant. Optimization-based iterative reconstruction approaches, e.g., compressive sensing provide ways to reduce the radiation exposure, without sacrificing image quality. However, the computational requirement such algorithms is much higher than that of the conventional Filtered Back Projection (FBP) reconstruction algorithm. This paper describes an FPGA implementation of one important iterative kernel called EM, which is the major computation kernel of a recent EM+TV reconstruction algorithm. We show that a hybrid approach (CPU+GPU+FPGA) can deliver a better performance and energy efficiency than GPU-only solutions, providing 13X boost of throughput than a dual-core CPU implementation.

A Hybrid Architecture for Compressive Sensing 3-D CT Reconstruction

The radiation dose associated with computerized tomography (CT) is significant. Compressive sensing (CS) methods provide mathematical approaches to reduce the radiation exposure without sacrificing reconstructed image quality. However, the computational requirements of these algorithms is much higher than conventional image reconstruction approaches such as filtered back projection (FBP). This paper describes a new compressive sensing 3-D image reconstruction algorithm based on expectation maximization and total variation, termed EM+TV, and also introduces a promising hybrid architecture implementation for this algorithm involving the combination of a CPU, GPU, and FPGA. An FPGA is used to speed up the major computation kernel (EM), and a GPU is used to accelerate the TV operations. The performance results indicate that this approach provides lower energy consumption and better reconstruction quality, and illustrates an example of the advantages that can be realized through domain-specific computing.

EM+TV based reconstruction for cone-beam CT with reduced radiation

Computerized tomography (CT) plays a critical role in modern medicine. However, the radiation associated with CT is significant. Methods that can enable CT imaging with less radiation exposure but without sacrificing image quality are therefore extremely important. This paper introduces a novel method for enabling image reconstruction at lower radiation exposure levels with convergence analysis. The method is based on the combination of expectation maximization (EM) and total variation (TV) regularization. While both EM and TV methods are known, their combination as described here is novel. We show that EM+TV can reconstruct a better image using much fewer views, thus reducing the overall dose of radiation. Numerical results show the efficiency of the EM+TV method in comparison to filtered backprojection and classic EM. In addition, the EM+TV algorithm is accelerated with GPU multicore technology, and the high performance speed-up makes the EM+TV algorithm feasible for future practical CT systems.

Efficient Video Coding Using Legacy Algorithmic Approaches

We show that for high bit rates, a video coding algorithm using a suitable combination of the QM coder and on other methods first published over 20 years ago can deliver video quality rivaling that of H.264 at lower complexity. This has implications both technically, since encoders built using these methods can be more power efficient, and commercially, given the complex licensing and intellectual property issues that accompany newer coding methods such as H.264 and MPEG-4. The methods described in this paper are the basis for the recent decision of the MPEG standards group to begin work on what is referred to as the “Type-1 Video Coding” standard, which, in addition to aiming for high coding efficiency, is intended to minimize royalty issues.

Adaptive Frequency Weighting for High-Performance Video Coding

The benefits of utilizing the spatial-frequency sensitivity of the human visual system in video coding have been known for decades and are incorporated into video standards, such as MPEG2, MPEG4, and H.264/AVC, as well as into still image-coding standards, such as JPEG and JPEG-2000. However, all of these standards utilize frequency weighting that is adaptive only at the picture level. In contrast with such picture-level adaptive frequency-weighting algorithms, we describe a coding approach in which a macroblock-level adaptive frequency-weighting (MBAFW) algorithm is used. We show that for the bitrate, quality levels, and resolutions typically associated with high-definition video, MBAFW can lead to overall bitrate reductions typically in the 5%-9% range over the methods used in state-of-the-art video-coding standards.

Overview of AVS broadcasting standard for high definition video

With rapid growth of HD video applications, AVS started HD video profile standardization process in March 2008 and finalized in September 2009. In the late of year 2011, the broadcasting industry standard for HD video compression was developed based on this work and has been recently completed and approved. It uses state-of-the-art coding tools and provides enhanced coding efficiency tools for high-definition video application. In this paper, an overview of this standard is provided, including the highlights of the capabilities of the new features tools. Some comparisons with the existing standards are also provided.

Parallel fast inter mode decision for H.264/AVC encoding

For H.264/AVC encoding, the mode selection process consumes a large proportion of the overall computation. To reduce this burden, various fast mode decision algorithms have been proposed. The current fast mode decision algorithms usually exploit the relationship among the coding modes and use the context-based approach to reduce the number of modes to be checked for both intra coding and inter coding. The parallel capacity of hardware architectures are also taken into consideration. However, almost all the parallel fast mode decision designs are focusing on intra coding. In this paper, a hardware friendly parallel fast inter mode decision method is proposed. With the proposed method, the inter mode decision can be conducted efficiently in one pass and significant encoding speedup can be achieved with negligible coding efficiency loss. Moreover, the proposed method can be easily mapped to hardware architecture which can be used for the real-time video encoding.

Quantization Matrix Coding for High Efficiency Video Coding

Quantization matrix (QM) has been adopted in image coding standards such as JPEG and JPEG-2000, as well as in video standards such as MPEG2, MPEG4 and H.264/AVC. QM can improve the subjective quality through frequency weighting on different frequency coefficients. In the latest high efficiency video coding (HEVC) standard, the quantization block sizes can go up to 32x32. To apply the frequency weighting techniques to HEVC, it needs multiple sizes (4x4, 8x8, 16x16 and 32x32) QMs. The bits to signal the multiple matrices will result in a huge overhead. In this paper, a predictive coding method for the quantization matrix is proposed. The bits consumption for QMs can be reduced significantly. Experimental results show that the proposed method is 28x times efficient (96.4% bit saving) than the quantization matrix compression method used in H.264/AVC. Moreover, the proposed method will only introduce negligible complexity on encoder and decoder.

Macroblock-Level Adaptive Frequency Weighting

In the perceptual video coding scheme, the properties of the human visual system (HVS) are usually used to improve the subjective quality. The frequency sensitivity is one of the most important properties of HVS and related techniques have been adopted in many video coding standards, such as MPEG2 and H.264/AVC High Profile. However, the frequency weighting algorithms used in those standards are all in picture level. In this paper, a novel macroblock (MB)-level adaptive frequency weighting algorithm is proposed. Compared with the picture-level adaptive frequency weighting algorithms, the proposed algorithm can select different frequency weighting strategies and quantization matrices for each MB. The experimental results show that the proposed MBAFW algorithm can improve the subjective quality significantly.

An adaptive covariance-based edge diffusion image enlargement method

We discuss image or video enlargement methods aimed at computationally constrained environments. Traditional enlargement algorithms such as linear or cubic interpolation have been applied in many applications. However the performance of these approaches is limited by artifacts such as blurring and jagged edges. More sophisticated iterative and learning-based algorithms have been proposed to address these issues, but they typically involve very high computational complexity. We present an adaptive covariance-based edge diffusion (ACED) image enlargement method that offers both good performance and low complexity. Different from other edge-directed interpolation algorithms, the proposed method uses combination of a novel edge-directed judgment which can choose different spread templates adaptively to estimate local covariance coefficients and edge diffusion to reduce artifacts. Experimental results show that the proposed method gives performs well both in terms of subjective quality as well as objective measures.