Object detection is an essential and expensive process in many computer vision systems. Standard off-the-shelf embedded processors are hard to achieve performance-power balance for implementation of object detection applications. In this work, we explore an Application Specific Instruction set Processor (ASIP) for object detection using Histogram of Oriented Gradients (HOG) feature. Algorithm simplifications are adopted to reduce memory bandwidth requirements and mathematical complexity without losing reliability. Also, parallel histogram generation and on-the-fly Support Vector Machine (SVM) calculation architecture are employed to reduce the necessary cycle counts. The HOG algorithm on the proposed ASIP was accelerated by a factor of 63x compared to the pure software implementation. The ASIP was synthesized for a standard 90nm CMOS library, with a silicon area of 1.31mm2 and 47.8mW power consumption at a 200MHz frequency. Our object detection processor can achieve 42 frames-per-second (fps) on VGA video. The evaluation and implementation results show that the proposed ASIP is both area-efficient and power-efficient while being competitive with commercial CPUs/DSPs. Furthermore, our ASIP exhibits comparable performance even with hard-wire designs.

Orientation field (OF) estimation is a fundamental process in fingerprint authentication systems. In this paper, a novel binary pattern based low-cost OF estimation algorithm is proposed. The new method consists of two modules. The first is block-level orientation estimation and averaging in vector space by pixel level orientation statistics. The second is orientation quantization and smoothing. In the second module, the continuous orientation is quantized into fixed orientations with sufficient resolution (interval between fixed orientations). An effective smoothing scheme on the quantized orientation space is also proposed. The proposed algorithm is capable of stably processing poor-quality fingerprint images and is validated by tests conducted on an adaptive OF matching scheme. The proposed algorithm is also implemented into a fingerprint System on Chip (SoC) to comfirm that it satisfies the strict requirements of embedded system.

It is theoretically impossible to restore the original fingerprint image from a sequence of line images captured by a line sensor. However, in this paper we propose a unique fingerprint-image-generation algorithm, which derives fingerprint images from sequences of line images captured at different swipe speeds by the line sensor. A continuous image representation, called trajectory, is used in modeling distortion of raw fingerprint images. Sequences of line images captured from the same finger are considered as sequences of points, which are sampled on the same trajectory in N-dimensional vector space. The key point here is not to reconstruct the original image, but to generate identical images from the trajectory, which are independent of the swipe speed of the finger. The method for applying the algorithm in a practical application is also presented. Experimental results on a raw fingerprint image database from a line sensor show that the generated fingerprint images are independent of swipe speed, and can achieve remarkable matching performance with a conventional minutiae matcher.

A novel binary line-pattern algorithm for embedded fingerprint authentication system is introduced in this paper. In this algorithm, each line-pattern is a one-dimension binary matrix that describes the alternation pattern of ridge and valley in fingerprint image. Two parallel lines or two cross lines in a certain scope make up related line-pattern pair. Several such line-pattern pairs at different parts of a fingerprint image can describe another intrinsic feature besides traditional minutiae feature. Experimental results showed this algorithm was not only efficient but also effective. Furthermore, a hybrid fingerprint match scheme is also introduced in this paper. It has the following features: (i) minutiae matching is firstly carried out to calculate the similarity score between the query fingerprint and the template fingerprint, and moreover, the translation and rotation parameters are obtained at the same time; (ii) line-pattern algorithm is immediately performed based on the parameters obtained after minutiae matching to get another similarity score; (iii) the final matching score is the combination of the minutiae matching score and the line-pattern matching score. Experiments were conducted on the FVC2002 database and our private database respectively. Both of the results were inspiring. In detail, at the same FAR value, the FRR of this hybrid match algorithm is to be 2-8% lower than only minutiae-based matching algorithm.

Object detection is at the heart of nearly all the computer vision systems. Standard off-the-shelf embedded processors are hard to meet the trade-offs among performance, power consumption and flexibility required by object detection applications. Therefore, this paper presents an Application Specific Instruction set Processor (ASIP) for object detection using AdaBoost-based learning algorithm with Haar-like features as weak classifiers. Algorithm optimizations are employed to reduce memory bandwidth requirements without losing reliability. In the proposed ASIP, Single Instruction Multiple Data (SIMD) architecture is adopted for fully exploiting data-level parallelism inherent to the target algorithm. With adding pipeline stages, application-specific hardware components and custom instructions, the AdaBoost algorithm is accelerated by a factor of 13.7x compared to the optimized pure software implementation. Compared with ARM946 and TMS320C64+, our ASIP shows 32x and 7x better throughput, 10x and 224x better area efficiency, 6.8x and 18.8x better power efficiency, respectively. Furthermore, compared to hard-wired designs, evaluation results show an advantage of the proposed architecture in terms of chip area efficiency while maintain a reliable performance and achieve real-time object detection at 32fps on VGA video.

In this paper, we will show some significant results of the routability analysis of bit-serial pipeline datapath designs based on Rent's rule and Donath's observation. Our results show that all of the tested bit-serial benchmarks have Rent exponent of below 0.4, indicating that the average wiring length of the circuit is expected to be independent of the circuit size. This study provides some important implications on the silicon utilization and time-area efficiency of bit-serial pipeline circuits on FPGAs and ASICs.

In this paper, a systematic method which generates the micro-architecture of Application Specific Instruction Processor (ASIP) is proposed. Different from previous works, the data path and control path are generated from the instruction sequence which is generated by translating the compiled assembly code. A graphical representation method called Register Transfer Graph (RTG) is introduced to describe the micro-operations of instruction sequence. To achieve high performance, we perform micro-operation level scheduling which dynamically assigns the micro-operations of instruction sequence to the control steps. By transforming the architecture using synthesis parameters, design space is explored more extensively. Connection cost is minimized by removing the inefficient data transfer paths.

In Internet of Things (IoT) applications, system-on-chip (SoCs) with embedded processors are widely used. As an embedded processor, RISC-V, which is license-free and has an extensible instruction set, is receiving attention. However, designing such embedded processors requires an enormous effort to achieve a highly efficient microarchitecture in terms of performance, power consumption, and circuit area, as well as the design verification of running complex software, including modern operating systems such as Linux. In this paper, we propose a method for directly describing the RTL structure of a pipelined RISC-V processor with cache memories, a memory management unit (MMU), and an AXI bus interface using the C++ language. This pipelined processor C++ model serves as a functional simulator of the complete RISC-V core, whereas our C2RTL framework translates the processor C++ model into a cycle-accurate RTL description in the Verilog-HDL and RTL-equivalent C model. Our processor design methodology using the C2RTL framework is unique compared to other existing methodologies because both the simulation and RTL models are derived from the same C++ source, which greatly simplifies the design verification and optimization processes. The effectiveness of our design methodology is demonstrated on a RISC-V processor that runs Linux OS on an FPGA board, achieving a significantly short simulation time of the original C++ processor model and RTL-equivalent C model in comparison to a commercial RTL simulator.

This paper proposes a designing algorithm for quadrilateral recursive filters which consist of four quarter-plane filters in the four quadrants. This can realize a perfect zero-phase filtering which is essential for image processing. Furthermore, several parallel processing algorithms capable of performing under very high parallel efficiency are developed on line-connected and mesh-connected processor arrays. By these proposals, the advantage of two-dimensional non-causal zero-phase recursive digital filters is made clear.

This paper firstly presents a processor design with Derivative ASIP approach. The architecture of processor is designed by making use of a well-known embedded processor's instruction-set as a base architecture. To improve its performance, the architecture is enhanced with more hardware resources such as registers, interfaces and instruction extensions which might achieve target specifications. Secondly, a new approach for retargeting compiler by means of assembly converter tool is proposed. Our retargeting approach is practical because it is performed by the assembly converter tool with a simple configuration file and independent from a base compiler. With our proposed approach, both architecture flexibility and a good quality of assembly code can be obtained at once. Compared to other compilers, experiments show that our approach capable of generating code as high efficiency as its base compiler and the developed ASIP results in better performance than its base processor.

Emerging image and video applications and conventional MPSoC architectures encounter drastically increasing performance and flexibility requirements. In order to display high quality images, large amount of image processing needs to be carried out. These image processing algorithms are nonstandard and vary case by case, and it is difficult to achieve real time processing by using general purpose processors or DSP. In this paper, we present two reconfigurable Application Specific Instruction-set Processors (ASIP) which can perform several image processing algorithms by using the same processor architecture. These ASIPs can achieve performance similar to DSP; meanwhile, while the area is considerably smaller than DSP and slightly bigger than conventional RISC processor. 1D ASIP can perform 16 times higher compared to a RISC processor, and 2D ASIP can perform 3 to 7 times higher compared to a RISC processor.

This paper presents a Multiprocessor System-on-Chips (MPSoC) architecture used as an execution platform for the new C-language based MPSoC design framework we are currently developing. The MPSoC architecture is based on an existing SoC platform with a commercial RISC core acting as the host CPU. We extend the existing SoC with a multiprocessor-array block that is used as the main engine to run parallel applications modeled in our design framework. Utilizing several optimizations provided by our compiler, an efficient inter-communication between processing elements with minimum overhead is implemented. A host-interface is designed to integrate the existing RISC core to the multiprocessor-array. The experimental results show that an efficacious integration is achieved, proving that the designed communication module can be used to efficiently incorporate off-the-shelf processors as a processing element for MPSoC architectures designed using our framework.

Narrow swipe sensor based systems have drawn more and more attention in recent years. However, the size of captured image is significantly smaller than that obtained from the traditional area fingerprint sensor. Under this condition the available minutiae number is also limited. Therefore, only employing minutiae with the standard associated feature can hardly achieve high verification accuracy. To solve this problem, we present a novel Hybrid Minutiae Descriptor (HMD) which consists of two modules. The first one: Minutiae Ridge-Valley Orientation Descriptor captures the orientation information around minutia and also the trace points located at associated ridge and valley. The second one: Gabor Binary Code extracts and codes the image patch around minutiae. The proposed HMD enhances the representation capability of minutiae feature, and can be matched very efficiently. Experiments conducted over public databases and the database captured by the narrow swipe sensor show that this innovative method gives rise to significant improvements in reducing FRR (False Reject Rate) and EER (Equal Error Rate).

Image processing engine is crucial for generating high quality images in video system. As Application Specific Integrated Circuit (ASIC) is dedicated for specific standards, Application Specific Instruction-set Processor (ASIP) which provides high flexibility and high performance seems to have more advantages in supporting the nonstandard pre/post image processing in video system. In our previous work, we have designed some ASIPs that can perform several image processing algorithms with a reconfigurable datapath. ASIP is as efficient as DSP, but its area is considerably smaller than DSP. As the resolution of image and the complexity of processing increase, the performance requirement also increases accordingly. In this paper, we presents a novel multi ASIP based image processing unit (IPU) which can provide sufficient performance for the emerging very-high-resolution applications. In order to provide a high performance image processing engine, we propose several new techniques and architecture such as multi block-pipes architecture, pixel direct transmission and boundary pixel write-through. Multi block-pipes architecture has flexible scalability in supporting a various ranges of resolution, which ranges from low resolution to high resolution. The boundary pixel write-through technique provides high efficient parallel processing, and pixel direct transmission technique is implemented in each ASIP to further reduce the data transmission time. Cycle-accurate SystemC simulations are performed, and the experimental results show that the maximum bandwidth of the proposed communication approach can achieve up to 1580 Mbyte/s at 400 MHz. Moreover, communication overhead can be reduced about a maximum of 88% compared to our previous works.

In this paper, a LSI design for video encoder and decoder for H.263+ video compression is presented. LSI operates under clock frequency of 27 MHz to compress QCIF (176144 pixels) at the frame rate of 30 frame per second. The core size is 4.6 4.6 mm2 in a 0.35 µm process. The architecture is based on bus connected heterogeneous dedicated modules, named as System-MSPA architecture. It employs the fast and small-chip-area dedicated modules in lower level and controls them by employing the slow and flexible programmable device and an external DRAM. Design results in success to achieve real time encoder in quite compact size without losing flexibility and expand ability. Real time emulation and easy test capability with external PC is also implemented.

As a global feature of fingerprint patterns, the Orientation Field (OF) plays an important role in fingerprint recognition systems. This paper proposes a fast binary pattern based orientation estimation with nearest-neighbor search, which can reduce the computational complexity greatly. We also propose a classified post processing with adaptive averaging strategy to increase the accuracy of the estimated OF. Experimental results confirm that the proposed method can satisfy the strict requirements of the embedded applications over the conventional approaches.

We propose a low cost and stand-alone platform-based SoC for H.264/AVC decoder, whose target is practical mobile applications such as a handheld video player. Both low cost and stand-alone solutions are particularly emphasized. The SoC, consisting of RISC core and decoder core, has advantages in terms of flexibility, testability and various I/O interfaces. For decoder core design, the proposed H.264/AVC coprocessor in the SoC employs a new block pipelining scheme instead of a conventional macroblock or a hybrid one, which greatly contribute to reducing drastically the size of the core and its pipelining buffer. In addition, the decoder schedule is optimized to block level which is easy to be programmed. Actually, the core size is reduced to 138 KGate with 3.5 kbyte memory. In our practical development, a single external SDRAM is sufficient for both reference frame buffer and display buffer. Various peripheral interfaces such as a compact flash, a digital broadcast receiver and a LCD driver are also provided on a chip.

An entropy decoding engine plays an important role in modern multimedia decoders. Previous researches that focused on the decoding performance paid a considerable attention to only one parameter such as the data parsing speed, but they did not consider the performance caused by a table configuration time and memory size. In this paper, we developed a novel method of entropy decoding based on the two step group matching scheme. Our approach achieves the high performance on both data parsing speed and configuration time with small memory needed. We also deployed our decoding scheme to implement an entropy decoding processor, which performs operations based on normal processor instructions and VLD instructions for decoding variable length codes. Several extended VLD instructions are prepared to increase the bitstream parsing process in modern multimedia applications. This processor provides a solution with software flexibility and hardware high speed for stand-alone entropy decoding engines. The VLSI hardware is designed by the Language for Instruction Set Architecture (LISA) with 23 Kgates and 110 MHz maximum clock frequency under TSMC 0.18 µm technology. The experimental simulations revealed that proposed processor achieves the higher performance and suitable for many practical applications such as MPEG-2, MPEG-4, H.264/AVC and AAC.

The fingerprint verification system is widely used in mobile devices because of fingerprint's distinctive features and ease of capture. Typically, mobile devices utilize small sensors, which have limited area, to capture fingerprint. Meanwhile, conventional fingerprint feature extraction methods need detailed fingerprint information, which is unsuitable for those small sensors. This paper proposes a novel fingerprint verification method for small area sensors based on deep learning. A systematic method combines deep convolutional neural network (DCNN) in a Siamese network for feature extraction and XGBoost for fingerprint similarity training. In addition, a padding technique also introduced to avoid wraparound error problem. Experimental results show that the method achieves an improved accuracy of 66.6% and 22.6% in the FingerPassDB7 and FVC2006DB1B dataset, respectively, compared to the existing methods.

In this paper, we present our work on the design of a new FPGA architecture targeted for high-performance bit-serial pipeline datapath. Bit-parallel systems require large amount of routing resource which is especially critical in using FPGAs. Their device utilization and operation frequency become low because of large routing penalty. Whereas bit-serial circuits are very efficient in routing, therefore are able to achieve a very high logic utilization. Our proposed FPGA architecture is designed taking into account the structure of bit-serial circuits to optimize the logic and routing architecture. Our FPGA guarantees near 100% logic utilization with a straightforward place and route tool due to high routability of bit-serial circuits and simple routing interconnect architecture. The FPGA chip core which we designed consists of around 200k transistors on 3.5 mm square substrate using 0.5 µm 2-metal CMOS process technology.