Convolution-based Filtering Research Articles

Learning graph-based features for relief patterns classification on mesh manifolds

Relief patterns represent a surface characteristic that can be seen as the 3D counterpart of the texture concept in 2D images. Such characteristic is well distinct from the 3D object shape but represents a good information source to recognize the object itself. The majority of state-of-the-art techniques for 2D images rely on convolution-based filtering so, the idea of extending such techniques to the mesh manifold domain is quite intriguing as much as challenging. In this paper, we propose a novel approach based on Graph Neural Networks for 3D mesh relief pattern classification. To this end, we designed a bi-level architecture that learns on data structures computed thanks to a mesh resampling algorithm that allows us to represent local surface patches uniformly, while keeping a consistent points order. The local mesh structures are represented by SpiderPatches, that aim to capture local features of the 3D mesh surface, providing a fine-grained rich representation of the relief patterns; global structures are instead captured by MeshGraphs, whose nodes are SpiderPatches, representing the mesh at a macroscopic level. We tested our architecture using SpiderPatches and MeshGraphs on the original meshes of the SHREC’17 and SHREC’20 relief patterns track datasets, showing superior performance to that reported in the literature using a comparable experimental setting.

A Sparse CNN Accelerator for Eliminating Redundant Computations in Intra- and Inter-Convolutional/Pooling Layers

Neural network pruning, which can be divided into unstructured pruning and structured pruning strategies, has been proven to be an efficient method to substantially reduce the number of computations of convolutional neural networks (CNNs). However, it remains difficult to combine the advantages of these two pruning strategies. This article proposes a high-performance accelerator for unstructured sparse CNNs. First, a convolution-based filter selection and clustering method (FSCM) is proposed to reorder unstructured sparse filters into uniform-size dense filters, eliminating redundant computations in the convolutional layers while maintaining a regular structure. In addition, a convolution and pooling calculation method (CPCM) is presented to reduce interlayers’ computational redundancy. Third, a hardware accelerator with high digital signal processing (DSP) efficiency is designed to take advantage of FSCM and CPCM. The proposed accelerator is implemented on a Xilinx XCVU9P platform at 300 MHz. With different CNN configurations, the performance and DSP efficiency for LeNet, AlexNet, and VGG16 are 1490.53 GOPS and 3.882, 913.85 GOPS and 1.785, and 862.16 GOPS and 1.68, respectively. Compared to previous field-programmable gate array (FPGA) and application-specific integrated circuit (ASIC) designs, the accelerator achieves up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$15.43\times $ </tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$7.73\times $ </tex-math></inline-formula> speedup in terms of performance and DSP efficiency.

Convolution operations for relief-pattern retrieval, segmentation and classification on mesh manifolds

Abstract Relief patterns represent a surface characteristic that is well distinct from the 3D object shape. They can be seen as the 3D counterpart of the texture concept in the 2D images. A large part of texture analysis, in 2D image state-of-the-art, relies on some convolution-based filtering. Thus, the idea of extending such techniques to the mesh manifold domain is quite natural. Nevertheless, defining a convolution operator on a mesh manifold is not straightforward. In this paper, we propose two frameworks, namely, Mesh-Grid and Mesh-Convolution, to apply discrete and continuous filters directly on the mesh. We tested Mesh-Grid and Mesh-Convolution in the task of geometric texture retrieval, providing, to the best of our knowledge, the first results on the SHREC’18 dataset. Then, our convolution operator revealed to be effective also in the task of relief pattern classification on the SHREC’17 dataset, outperforming the state-of-the-art results. Finally, we propose a geometric texture segmentation approach to support manual annotation on large datasets, which revealed to be effective.

Partial Response Signaling for Circular Convolution Based Filter Bank Multi-Carrier System Adopting Quadrature Amplitude Modulation

A circular convolution-based filter bank multi-carrier system adopting quadrature amplitude modulation (CQAM-FBMC) that utilizes partial response signaling (PRS), referred to as the PRS-CQAM-FBMC system, is proposed and analyzed. The proposed PRS-CQAM-FBMC system has fully overlapped FBMC symbols without the time shifts and the transition time through the circular convolution filter banks in the time domain and the improved localization property by the PRS of the poorly localized QAM-FBMC symbols in the frequency domain. Therefore, the PRS-CQAM-FBMC system can achieve higher spectral efficiency than the conventional QAM-FBMC system in short-packet transmission scenarios. The analyses and simulation results show that the symbol error rate performances of the PRS-CQAM-FBMC system without cyclic prefix are robust in additive white Gaussian channels and multi-path fading channels. Moreover, the PRS-CQAM-FBMC system with the improved localization property by PRS can further provide enhanced spectral efficiency by more than 10%.

Multiplier-Less Stream Processor for 2D Filtering in Visual Search Applications

A new 2D convolution-based filter is presented, which is specifically designed to improve visual search applications. It exploits a new radix-3 partitioning method of integer numbers, derived from the weight partition theory, which allows substituting multipliers with simplified floating point (FP) adders, working on 32-b FP filter coefficients. The memory organization allows elaborating the incoming data in raster scan order, as those directly provided by an acquisition source, without frame buffers and additional aligning circuitry. Compared with the existent literature, built around conventional arithmetic circuitry, the proposed design achieves state-of-the-art performances in the reduction of the mapped physical resources and elaboration velocity, achieving a critical path delay of about 4.5 ns both with a Xilinx Virtex-7 field-programmable gate array and CMOS 90-nm std_cells.

A modified listless strip based SPIHT for wireless multimedia sensor networks

Set Partitioning In Hierarchical Tree (SPIHT) is considered one of the most important algorithms for reducing the size of the vision data collected by the sensor node within wireless multimedia sensor network (WMSN). The traditional SPIHT algorithm suffers from image coders complexity due to large memory requirement. This is an essential problem for the implementation on limited resource environments such as WMSN. The main objective of this paper is to introduce a listless pipelined strip based SPIHT for WMSN to reduce system complexity and minimize processing time and memory usage. The proposed algorithm is implemented using discrete wavelet transform (DWT) lifting-based instead of DWT convolution-based filter. The experimental results show the superiority of the proposed algorithm in terms of peak signal-to-noise ratio (PSNR) which reaches 1 dB for all bit rates. In addition, the memory requirement is reduced to 71% with 27% of energy saving.

Numerical identification of a nonlinear diffusion coefficient by discrete mollification

The discrete mollification method is a convolution-based filtering procedure suitable for the regularization of ill-posed problems. Combined with explicit space-marching finite difference schemes, it provides stability and convergence for a variety of coefficient identification problems in linear parabolic equations. In this paper, we extend such a technique to identify some nonlinear diffusion coefficients depending on an unknown space dependent function in one dimensional parabolic models. For the coefficient recovery process, we present detailed error estimates and to illustrate the performance of the algorithms, several numerical examples are included.

Pipelined Discrete Wavelet Transform Architecture Scanning Dual Lines

A new discrete wavelet transform (DWT) architecture is proposed to realize a memory-efficient 2-D DWT processor. The proposed DWT processor conforms to dual-line scanning to remove the transpose buffer. In the previous single-line DWT architectures, the transpose buffer size is proportional to the row size of the image. The conventional dual-line DWT architecture is constructed by using the convolution-based filter structure and replicates registers to alternatively deal with two lines, resulting in a long delay, as well as a number of operators and registers. The proposed architecture is based on the lifting-based DWT to embed the additional registers in the middle of the DWT operation. In addition, the computation topology is optimized for the proposed dual-line DWT architecture to achieve almost the same hardware cost and critical path as the single-line DWT architecture.