Search

In this brief, we consider the problem of descriptors construction for the task of content-based image retrieval using deep neural networks. The idea of neural codes, based on fully connected layers’ activations, is extended by incorporating the information contained in convolutional layers. It is known that the total number of neurons in the convolutional part of the network is large and the majority of them have little influence on the final classification decision. Therefore, in this brief, we propose a novel algorithm that allows us to extract the most significant neuron activations and utilize this information to construct effective descriptors. The descriptors consisting of values taken from both the fully connected and convolutional layers perfectly represent the whole image content. The images retrieved using these descriptors match semantically very well to the query image, and also, they are similar in other secondary image characteristics, such as background, textures, or color distribution. These features of the proposed descriptors are verified experimentally based on the IMAGENET1M dataset using the VGG16 neural network. For comparison, we also test the proposed approach on the ResNet50 network. [ABSTRACT FROM AUTHOR]

The pulse-coupled neural network (PCNN) model is a third-generation artificial neural network without training that uses the synchronous pulse bursts of neurons to process digital images, but the lack of in-depth theoretical research limits its extensive application. By analyzing the working mechanism of the PCNN, we present an expression for the fire-extinguishing time of neurons that fire in the second iteration and an expression for the firing time of neurons that extinguish in the second iteration. In addition, we find a phenomenon of the PCNN and name it mathematically coupled fire extinguishing. Based on the above analysis, we propose a new working mode for the PCNN, where the refiring of fire-extinguishing neurons is only allowed when all firing neurons are extinguished. We also work out the constraint conditions of the parameter settings under this mode. Furthermore, we analyze the relationship between the network parameters and mathematically coupled fire extinguishing, the coupling of neighboring neurons, and the convergence rate of the PCNN, respectively. In addition, we demonstrate the essential regularity of extinguished neuron in the PCNN and then propose an optimal parameter setting to achieve the best comprehensive performance of the PCNN. [ABSTRACT FROM AUTHOR]

The pantograph–catenary system is crucial to the transmission of electrical power from catenary lines to electrified trains. The occurrence of arcs could damage or interrupt railway operations. We propose a CNN-based model to detect arcs and recognize their magnitudes. First, we decompose the pantograph videos recorded by a camera fixed on China High-Speed train into continuous frames, and grayscale-process and segment those images to obtain an arc image set. Then, we divide the image set into training samples and test samples. And the training samples are further divided into three classifications labeled as 0, 1, 2, which are used to train CNN model. The accuracy of CNN trained result reaches 0.95, and the loss function converges to 0.083. Second, we use the trained network to detect arcs in the images of the test samples, and convert arc detection results to a time series of arc scores. Thus the occurrence of arcs and their magnitudes can be determined. Finally, we conduct experiments to compare our approach with other models. The results demonstrate the approach's high accuracy, robustness, and high speed, when dealing with images taken from unstable surroundings. It could be applied to other EMU models or environments with adjusted parameters. [ABSTRACT FROM AUTHOR]

Recent face detectors based on deep convolutional neural networks (DCNN) have substantially improved the detection performance in the wild. However, the detection speed is still the biggest bottleneck, which hinders the practical deployment of these face detectors on resource-limited platforms. To address this issue, the paper proposes a Proposal Pyramid Network (PPN) to generate face candidates extremely fast, which reduces the major computational complexity in cascaded DCNN face detectors. PPN is a lightweight fully convolutional network with multiple branches, which takes a single image as input and generates face proposals with different scales in terms of separate branches simultaneously. In this manner, it avoids the traditional image pyramid structure and thus achieves a very fast processing speed. To evaluate the effectiveness of our proposed method, a three-stage cascaded DCNN face detector is realized based on PPN. Extensive experimental results on several public benchmarks show that the proposed face detector achieves comparable accuracy to the-state-of-arts while the detection speed reaches 60 FPS with an i5 CPU, which significantly exceeds previous DCNN-based face detectors. (The source code is available at https://github.com/zhaofan0622/PPN). [ABSTRACT FROM AUTHOR]

The PSNR and MSE are the computationally simplest and thus most widely used measures for image quality, although they correlate only poorly with perceived visual quality. More accurate quality models that rely on processing on both the reference and distorted image are potentially difficult to integrate in time-critical communication systems where computational complexity is disadvantageous. This paper derives the concept of distortion sensitivity as a property of the reference image that compensates for a given computational quality model a potential lack of perceptual relevance. This compensation method is applied to the PSNR and leads to a local weighting scheme for the MSE. Local weights are estimated by a deep convolutional neural network and used to improve the PSNR in a computationally graceful distribution of computationally complex processing to the reference image only. The performance of the proposed estimation approach is evaluated on LIVE, TID2013 and CSIQ databases and shows comparable or superior performance compared to benchmark image quality measures. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) are powerful machine learning models that have become the state of the art in several problems in the areas of computer vision and image processing. Nevertheless, the knowledge of why and how these models present an impressive performance is still limited. There are visualization techniques that can help us to understand the inner working of neural networks. However, they have mostly been applied to classification models. In this paper, we evaluate the application of visualization methods to networks where the input and output are images of proportional dimensions. The results show that visualization brings visual cues associated with how these systems work, helping in their understanding and improvement. We use the knowledge obtained from the visualization of an image restoration CNN to improve the architecture’s efficiency with no significant degradation of its performance. [ABSTRACT FROM AUTHOR]

Highlights • We propose a new method for OLHCCR, which is entirely based on 1-D CNN. • Our method achieves 98.11% on ICDAR-2013 and 97.14% on IAHCCR-UCAS2016. • Our method is different from the 2-D CNN or the RNN method for OLHCCR. • Our model is smaller than the 2-D CNN model and is faster than the RNN model. Abstract For online handwritten Chinese character recognition (OLHCCR), it has become a popular choice to employ the 2-dimensional convolutional neural network (2-D CNN) in recognizing the extracted feature images or utilize the recurrent neural network (RNN) to classify handwriting trajectories directly. Instead, here we propose to apply the 1-dimensional CNN (1-D CNN), which, to our knowledge, is novel in the context of OLHCCR. Specifically, a 1-D CNN is engaged upon the sequential handwriting trajectories. Each output sequence is then averaged over time to form a fixed-size vector representation, upon which the final classification is made via a softmax operation. Compared with the 2-D CNN architecture, our approach is capable of delivering better results with a more compact model. This is achieved without adopting the computationally demanding techniques that are necessary when working with the 2-D CNN, including data augmentation and feature image extraction. Furthermore, our method attains a faster test time speed compared with the RNN, and this is more pronounced in processing long sequences. Empirically, our method yields the near-state-of-the-art accuracy of 98.11% on ICDAR 2013 competition dataset, and the state-of-the-art accuracy of 97.14% on in-air handwriting dataset IAHCC-UCAS2016, respectively. [ABSTRACT FROM AUTHOR]

To detect changed areas in multitemporal polarimetric synthetic aperture radar (SAR) images, this paper presents a novel version of convolutional neural network (CNN), which is named local restricted CNN (LRCNN). CNN with only convolutional layers is employed for change detection first, and then LRCNN is formed by imposing a spatial constraint called local restriction on the output layer of CNN. In the training of CNN/LRCNN, the polarimetric property of SAR image is fully used instead of manual labeled pixels. As a preparation, a similarity measure for polarimetric SAR data is proposed, and several layered difference images (LDIs) of polarimetric SAR images are produced. Next, the LDIs are transformed into discriminative enhanced LDIs (DELDIs). CNN/LRCNN is trained to model these DELDIs by a regression pretraining, and then a classification fine-tuning is conducted with some pseudolabeled pixels obtained from DELDIs. Finally, the change detection result showing changed areas is directly generated from the output of the trained CNN/LRCNN. The relation of LRCNN to the traditional way for change detection is also discussed to illustrate our method from an overall point of view. Tested on one simulated data set and two real data sets, the effectiveness of LRCNN is certified and it outperforms various traditional algorithms. In fact, the experimental results demonstrate that the proposed LRCNN for change detection not only recognizes different types of changed/unchanged data, but also ensures noise insensitivity without losing details in changed areas. [ABSTRACT FROM AUTHOR]

This paper proposes a new discriminative deep quantization hashing (DDQH) approach for large-scale face image retrieval by learning discriminative and compact binary codes. It jointly explores the discrete code learning, batch normalization quantization (BNQ) module, and end-to-end learning in one unified framework, which can guarantee the optimal compatibility of hash coding and feature learning. To learn multiscale and robust facial features, a deep network properly stacking several convolution-pooling layers and pooling layers is designed, and the facial features are obtained by fusing the outputs of the last convolutional layer and the last pooling layer. Besides, the prediction errors of the learned binary codes are minimized to learn discriminative binary codes of images. To obtain higher retrieval accuracies, a BNQ module is utilized to control quantization at a moderate level. Experiments are conducted on two widely used data sets, and the proposed DDQH method achieves encouraging improvements over some state-of-the-art hashing approaches. [ABSTRACT FROM AUTHOR]

According to the oversampling imaging characteristics, an infrared small target detection method based on deep learning is proposed. A 7-layer deep convolutional neural network (CNN) is designed to automatically extract small target features and suppress clutters in an end-to-end manner. The input of CNN is an original oversampling image while the output is a cluttersuppressed feature map. The CNN contains only convolution and non-linear operations, and the resolution of the output feature map is the same as that of the input image. The L1-norm loss function is used, and a mass of training data is generated to train the network effectively. Results show that compared with several baseline methods, the proposed method improves the signal clutter ratio gain and background suppression factor by 3 - 4 orders of magnitude, and has more powerful target detection performance. [ABSTRACT FROM AUTHOR]

In this paper, we build a multilabel image classifier using a general deep convolutional neural network (DCNN). We propose a novel objective function that consists of three parts, i.e., max-margin objective, max-correlation objective, and correntropy loss. The max-margin objective explicitly enforces that the minimum score of positive labels must be larger than the maximum score of negative labels by a predefined margin, which not only improves accuracies of the multilabel classifier, but also eases the threshold determination. The max-correlation objective can make the DCNN model learn a latent semantic space, which maximizes the correlations between the feature vectors of the training samples and their corresponding ground-truth label vectors projected into this space. Instead of using the traditional softmax loss, we adopt the correntropy loss from the information theory field to minimize the training errors of the DCNN model. The proposed framework can be end-to-end trained. Comprehensive experimental evaluations on Pascal VOC 2007 and MIR Flickr 25K multilabel benchmark data sets with four DCNN models, i.e., AlexNet, VGG-16, GoogLeNet, and ResNet demonstrate that the proposed objective function can remarkably improve the performance accuracies of a DCNN model for the task of multilabel image classification. [ABSTRACT FROM AUTHOR]

Fusion methods based on multi-scale transforms have become the mainstream of the pixel-level image fusion. However, most of these methods cannot fully exploit spatial domain information of source images, which lead to the degradation of image. This paper presents a fusion framework based on block-matching and 3D (BM3D) multi-scale transform. The algorithm first divides the image into different blocks and groups these 2D image blocks into 3D arrays by their similarity. Then it uses a 3D transform which consists of a 2D multi-scale and a 1D transform to transfer the arrays into transform coefficients, and then the obtained low- and high- coefficients are fused by different fusion rules. The final fused image is obtained from a series of fused 3D image block groups after the inverse transform by using an aggregation process. In the experimental part, we comparatively analyze some existing algorithms and the using of different transforms, e.g. non-subsampled Contourlet transform (NSCT), non-subsampled Shearlet transform (NSST), in the 3D transform step. Experimental results show that the proposed fusion framework can not only improve subjective visual effect, but also obtain better objective evaluation criteria than state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Visual and tactile sensing are complementary factors in the task of robotic grasping. In this paper, a grasp detection deep network is first proposed to detect the grasp rectangle from the visual image， then a new metric using tactile sensing is designed to assess the stability of the grasp. By means of this scheme, a THU grasp dataset， which includes the visual information, corresponding tactile and grasp configurations， is collected to train the proposed deep network. Experiments results have demonstrated that the proposed grasp detection deep networks outperform other mainstream approaches in a public grasp dataset. Furthermore， the grasp success rate can be improved significantly in real world scenarios. The trained model has also been successfully implemented in a new robotic platform to perform the robotic grasping task in a cluttered scenario. [ABSTRACT FROM AUTHOR]

The memristor has been extensively studied in electrical engineering and biological sciences as a means to compactly implement the synaptic function in neural networks. The cellular neural network (CNN) is one of the most implementable artificial neural network models and capable of massively parallel analog processing. In this paper, a novel memristive multilayer CNN (Mm-CNN) model is presented along with its performance analysis and applications. In this new CNN design, the memristor crossbar circuit acts as the synapse, which realizes one signed synaptic weight with a pair of memristors and performs the synaptic weighting compactly and linearly. Moreover, the complex weighted summation is executed in an efficient way with a proper design of Mm-CNN cell circuits. The proposed Mm-CNN has several merits, such as compactness, nonvolatility, versatility, and programmability of synaptic weights. Its performance in several image processing applications is illustrated through simulations. [ABSTRACT FROM AUTHOR]

In this paper, a dynamic delay interval (DDI) method is proposed to deal with the stability problem of neural networks with two delay components. This method extends the fixed interval of a time-varying delay to a dynamic one, which relaxes the restriction on upper and lower bounds of the delay intervals. Combining the reciprocally convex combination technique and Wirtinger integral inequality, the DDI method leads to some much less conservative delay-dependent stability criteria based on a linear matrix inequality for neural networks with two delay components. Furthermore, the criteria for the system with a single time-varying delay are provided. Some examples are given to illustrate the effectiveness of the obtained results. [ABSTRACT FROM AUTHOR]

This paper presents the design of radial basis function geometric bioinspired networks and their applications. Until now, the design of neural networks has been inspired by the biological models of neural networks but mostly using vector calculus and linear algebra. However, these designs have never shown the role of geometric computing. The question is how biological neural networks handle complex geometric representations involving Lie group operations like rotations. Even though the actual artificial neural networks are biologically inspired, they are just models which cannot reproduce a plausible biological process. Until now researchers have not shown how, using these models, one can incorporate them into the processing of geometric computing. Here, for the first time in the artificial neural networks domain, we address this issue by designing a kind of geometric RBF using the geometric algebra framework. As a result, using our artificial networks, we show how geometric computing can be carried out by the artificial neural networks. Such geometric neural networks have a great potential in robot vision. This is the most important aspect of this contribution to propose artificial geometric neural networks for challenging tasks in perception and action. In our experimental analysis, we show the applicability of our geometric designs, and present interesting experiments using 2-D data of real images and 3-D screw axis data. In general, our models should be used to process different types of inputs, such as visual cues, touch (texture, elasticity, temperature), taste, and sound. One important task of a perception–action system is to fuse a variety of cues coming from the environment and relate them via a sensor–motor manifold with motor modules to carry out diverse reasoned actions. [ABSTRACT FROM AUTHOR]

Multivariate mathematical morphology (MMM) aims to extend the mathematical morphology from gray scale images to images whose pixels are high-dimensional vectors, such as remote sensing hyperspectral images and functional magnetic resonance images (fMRIs). Defining an ordering over the multidimensional image data space is a fundamental issue MMM, to ensure that ensuing morphological operators and filters are mathematically consistent. Recent approaches use the outputs of two-class classifiers to build such reduced orderings. This paper presents the applications of MMM built on reduced supervised orderings based on lattice autoassociative memories (LAAMs) recall error measured by the Chebyshev distance. Foreground supervised orderings use one set of training data from a foreground class, whereas background/foreground supervised orderings use two training data sets, one for each relevant class. The first case study refers to the realization of the thematic segmentation of the hyperspectral images using spatial–spectral information. Spectral classification is enhanced by a spatial processing consisting in the spatial correction guided by a watershed segmentation computed by the LAAM-based morphological operators. The approach improves the state-of-the-art hyperspectral spatial–spectral thematic map building approaches. The second case study is the analysis of resting state fMRI data, working on a data set of healthy controls, schizophrenia patients with and without auditory hallucinations. We perform two experiments: 1) the localization of differences in brain functional networks on population-dependent templates and 2) the classification of subjects into each possible pair of cases. In this data set, we find that the LAAM-based morphological features improve over the conventional correlation-based graph measure features often employed in fMRI data classification. [ABSTRACT FROM PUBLISHER]

Salient object detection increasingly receives attention as an important component or step in several pattern recognition and image processing tasks. Although a variety of powerful saliency models have been intensively proposed, they usually involve heavy feature (or model) engineering based on priors (or assumptions) about the properties of objects and backgrounds. Inspired by the effectiveness of recently developed feature learning, we provide a novel deep image saliency computing (DISC) framework for fine-grained image saliency computing. In particular, we model the image saliency from both the coarse-and fine-level observations, and utilize the deep convolutional neural network (CNN) to learn the saliency representation in a progressive manner. In particular, our saliency model is built upon two stacked CNNs. The first CNN generates a coarse-level saliency map by taking the overall image as the input, roughly identifying saliency regions in the global context. Furthermore, we integrate superpixel-based local context information in the first CNN to refine the coarse-level saliency map. Guided by the coarse saliency map, the second CNN focuses on the local context to produce fine-grained and accurate saliency map while preserving object details. For a testing image, the two CNNs collaboratively conduct the saliency computing in one shot. Our DISC framework is capable of uniformly highlighting the objects of interest from complex background while preserving well object details. Extensive experiments on several standard benchmarks suggest that DISC outperforms other state-of-the-art methods and it also generalizes well across data sets without additional training. The executable version of DISC is available online: http://vision.sysu.edu.cn/projects/DISC. [ABSTRACT FROM PUBLISHER]

This paper presents a novel change detection approach for synthetic aperture radar images based on deep learning. The approach accomplishes the detection of the changed and unchanged areas by designing a deep neural network. The main guideline is to produce a change detection map directly from two images with the trained deep neural network. The method can omit the process of generating a difference image (DI) that shows difference degrees between multitemporal synthetic aperture radar images. Thus, it can avoid the effect of the DI on the change detection results. The learning algorithm for deep architectures includes unsupervised feature learning and supervised fine-tuning to complete classification. The unsupervised feature learning aims at learning the representation of the relationships between the two images. In addition, the supervised fine-tuning aims at learning the concepts of the changed and unchanged pixels. Experiments on real data sets and theoretical analysis indicate the advantages, feasibility, and potential of the proposed method. Moreover, based on the results achieved by various traditional algorithms, respectively, deep learning can further improve the detection performance. [ABSTRACT FROM PUBLISHER]

Cellular nonlinear/neural network (CNN) has been recognized as a powerful massively parallel architecture capable of solving complex engineering problems by performing trillions of analog operations per second. The memristor was theoretically predicted in the late seventies, but it garnered nascent research interest due to the recent much-acclaimed discovery of nanocrossbar memories by engineers at the Hewlett-Packard Laboratory. The memristor is expected to be co-integrated with nanoscale CMOS technology to revolutionize conventional von Neumann as well as neuromorphic computing. In this paper, a compact CNN model based on memristors is presented along with its performance analysis and applications. In the new CNN design, the memristor bridge circuit acts as the synaptic circuit element and substitutes the complex multiplication circuit used in traditional CNN architectures. In addition, the negative differential resistance and nonlinear current–voltage characteristics of the memristor have been leveraged to replace the linear resistor in conventional CNNs. The proposed CNN design has several merits, for example, high density, nonvolatility, and programmability of synaptic weights. The proposed memristor-based CNN design operations for implementing several image processing functions are illustrated through simulation and contrasted with conventional CNNs. Monte-Carlo simulation has been used to demonstrate the behavior of the proposed CNN due to the variations in memristor synaptic weights. [ABSTRACT FROM AUTHOR]

Benefiting from its effectiveness in subspace segmentation, low-rank representation (LRR) and its variations have many applications in computer vision and pattern recognition, such as motion segmentation, image segmentation, saliency detection, and semisupervised learning. It is known that the standard LRR can only work well under the assumption that all the subspaces are independent. However, this assumption cannot be guaranteed in real-world problems. This paper addresses this problem and provides an extension of LRR, named structure-constrained LRR (SC-LRR), to analyze the structure of multiple disjoint subspaces, which is more general for real vision data. We prove that the relationship of multiple linear disjoint subspaces can be exactly revealed by SC-LRR, with a predefined weight matrix. As a nontrivial byproduct, we also illustrate that SC-LRR can be applied for semisupervised learning. The experimental results on different types of vision problems demonstrate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Abstract: Spiking neural P systems with weights (WSN P systems, for short) are a class of distributed parallel computing devices inspired from the way neurons communicate by means of spikes. It has been proved that WSN P systems can generate/recognize Turing computable set of numbers (i.e., they are Turing universal as number generators/recognizers). In this work, we investigate the language generation power of WSN P systems, where the set of spike trains of halting computations of a given WSN P system constitutes the language generated by that system. Several relationships of the families of languages generated by WSN P systems with the family of finite languages and the family of regular languages are obtained. The family of recursively enumerable languages is characterized by projections of inverse-morphic images of languages generated by WSN P systems. [Copyright &y& Elsevier]

A new method for improving the storage capacity of associative memory models on a neural network is proposed. The storage capacity of the network increases in proportion to the network size in the case of random patterns, but, in general, the capacity suffers from correlation among memory patterns. Numerous solutions to this problem have been proposed so far, but their high computational cost limits their scalability. In this paper, we propose a novel and simple solution that is locally computable without any iteration. Our method involves XNOR masking of the original memory patterns with random patterns, and the masked patterns and masks are concatenated. The resulting decorrelated patterns allow higher storage capacity at the cost of the pattern length. Furthermore, the increase in the pattern length can be reduced through blockwise masking, which results in a small amount of capacity loss. Movie replay and image recognition are presented as examples to demonstrate the scalability of the proposed method. [ABSTRACT FROM AUTHOR]

In this paper, first, we investigate the CNN templates producing periodic patterns. Next, two image processing algorithms using periodic pattern formation in cellular neural networks are proposed. By applying one to a certain image, area having spatial uniformity is removed from the original image and only parts without spatial uniformity remain. Simulation results show that when this algorithm is applied to an image containing characters (texts) or a human face image, only characters or eyes and mouth are extracted. Also, by applying the other to texture image, periodicity is enhanced in area having spatial periodicity, and texture can be segmented. Simulation results show that when this algorithm is applied to image that consists of four kinds of textures, each texture area can be segmented. [ABSTRACT FROM AUTHOR]

In this paper, we show how a cellular nonlinear network with 1-bit weight programmability can be used for processing black and white image data. When using such a binary-programmable network, some templates need to be processed algorithmically, in other words, divided into subtasks that are processed consecutively. We classify templates into groups based on their properties and give guidelines as to how the division into subtasks (when applicable) is performed. A large collection of templates suitable for the proposed model is shown. We also describe one possible cell structure that realizes the binary-programmable model. The cell is modeled with Matlab and selected template simulations are shown. [ABSTRACT FROM AUTHOR]

The article discusses the growth in Artificial Intelligence (AI) technology poised for growth in embedded vision and image processing applications. Topics discussed include progress in terms of computational capacity, data accumulation, and improvements in mathematical tools; focus of AI on the ability to analyze images from vision systems; and use of tools such as embedOS from ARM.

This paper introduces the processing core of a full-custom mixed-signal CMOS chip intended for an active-contour-based technique, the so-called pixel-level snakes (PLS). Among the different parameters to optimize on the top-down design flow our methodology is focused on area. This approach results in a single-instruction-multiple-data chip implemented by a discrete-time cellular neural network with a correspondence between pixel and processing element. This is the first prototype for PLS; an integrated circuit with a 9 × 9 resolution manufactured in a 0.25-μm CMOS ST Microelectronics technology process. Awaiting for experimental results, HSPICE simulations prove the validity of the approach introduced here. [ABSTRACT FROM AUTHOR]

Today, with 0.18-μm technologies mature and stable enough for mixed-signal design with a large variety of CMOS compatible optical sensors available and with 0.09-μm technologies knocking at the door of designers, we can face the design of integrated systems, instead of just integrated circuits. In fact, significant progress has been made in the last few years toward the realization of vision systems on chips (VSoCs). Such VSoCs are eventually targeted to integrate within a semiconductor substrate the functions of optical sensing, image processing in space and time, high-level processing, and the control of actuators. The consecutive generations of ACE chips define a roadmap toward flexible VSoCs. These chips consist of arrays of mixed-signal processing elements (PEs) which operate in accordance with single instruction multiple data (SIMD) computing architectures and exhibit the functional features of CNN Universal Machines. They have been conceived to cover the early stages of the visual processing path in a fully-parallel manner, and hence more efficiently than DSP-based systems. Across the different generations, different improvements and modifications have been made looking to converge with the newest discoveries of neurobiologists regarding the behavior of natural retinas. This paper presents considerations pertaining to the design of a member of the third generation of ACE chips, namely to the so-called ACE16k chip. This chip, designed in a 0.35-μm standard CMOS technology, contains about 3.75 million transistors and exhibits peak computing figures of 330 GOPS, 3.6 GOPS/mm2 and 82.5 GOPS/W. Each PE in the array contains a reconfigurable computing kernel capable of calculating linear convolutions on 3 × 3 neighborhoods in less than 1.5 μs, imagewise Boolean combinations in less than 200 ns, imagewise arithmetic operations in about 5 μs, and CNN-like temporal evolutions with a time constant of about 0.5μs. Unfortunately,... [ABSTRACT FROM AUTHOR]

We propose a programmable architecture for a single instruction multiple data image processor that has its foundation on the mathematical framework of a simplicial cellular neural networks. We develop instruction primitives for basic Image processing operations and show examples of processing binary and gray scale images. Fabricated in deep submicron CMOS technologies, the complexity of the digital circuits and wiring in each cell is commensurate with pixel level processing. [ABSTRACT FROM AUTHOR]

This paper describes the electronic implementation of a four-layer cellular neural network architecture implementing two components of a functional model of neurons in the visual cortex: linear orientation selective filtering and half wave rectification. Separate ON and OFF layers represent the positive and negative outputs of two-phase quadrature Gabor-type filters, whose orientation and spatial-frequency tunings are electronically adjustable. To enable the construction of a multichip network to extract different orientations in parallel, the chip includes an address event representation (AER) transceiver that accepts and produces two-dimensional images that are rate encoded as spike trains. It also includes routing circuitry that facilitates point-to-point signal fan in and fan out. We present measured results from a 32 × 64 pixel prototype, which was fabricated in the TSMC0.25-μm process on a 3.84 by 2.54 mm die. Quiescent power dissipation is 3 mW and is determined primarily by the spike activity on the AER bus. Settling times are on the order of a few milliseconds. In comparison with a two-layer network implementing the same filters, this network results in a more symmetric circuit design with lower quiescent power dissipation, albeit at the expense of twice as many transistors. [ABSTRACT FROM AUTHOR]

The article discusses a study in which researchers at the online social networking service Facebook and Virginia Polytechnic Institute & State University mapped the areas of images that humans and neural-network-based artificial intelligence focused on to identify the differences.

Presents a study which presented a neural network-based method of image processing for determining the local composition and thickness of III-V semiconductors in high resolution electron microscope (HREM) images. Description of HREM images; Analysis of the neural network-based method; Experimental results; Conclusion.

Discusses the possibility of using neural networks to automate the vehicle-occupancy monitoring process using face detection and image recognition to eliminate the honor system for high-occupancy-vehicle (HOV) traffic lanes in Great Britain. Addition of HOV lanes to alleviate traffic congestion and encourage car pooling; Systems to enforce HOV regulations; Detection of faces within the windshield area.

Discusses the development of a hybrid analog/digital computer for image processing and machine vision applications. Features of cellular neural networks (CNN); Inventors of CNN; Advantages of the network.