Applications of Cellular Neural Networks to Image Understanding
Mariofanna Milanova
Sign up for access to the world's latest research
checkGet notified about relevant papers
checkSave papers to use in your research
checkJoin the discussion with peers
checkTrack your impact
Abstract
The Cellular Neural Networks (CNN) model is now a paradigm of ceIlular analog programmable multidimensional processor array with distributed local logic and memory. CNNs consist of many paraIlel analogue processors computing in real time. One desirable feature is that these processors arranged in a two dimensional grid, only have local connections, which lend themselves easily to VLSI implementations. The connections between these processors are determined by a cIoning template, which describes the strength of nearest-neighbour interconnections. The cloning templates are spaceinvariant, meaning that a11 the processors have the same relative connections. In this paper first we describe the architecture of CNN. Next, a new application of CNN using them for the 3D scene analysis is studied. .
Related papers
Cellular neural networks: a paradigm for nonlinear spatio-temporal processing
IEEE Circuits and Systems Magazine, 2001
The paradigm of Cellular Neural Networks (CNNs)is going to achieve a complete maturity. In fact, from a methodological point of view, important results on their digitally programmable analog dynamics have been gained, completed with thousands of application routines. This has encouraged the spreading of a great number of applications in the most different disciplines. Moreover, their structure, tailor made for VLSI realization, has led to the production of some chip prototypes that, embedded in a computational infrastructure, have produced the first analogic cellular computers. This completes the framework and makes it possible to realize complex spatio-temporal and filtering tasks on a time scale of microseconds. In this paper some sketches on the main aspects of CNNs, from the formal to the hardware prototype point of view, are presented together with some appealing applications to illustrate complex image, visual and spatio-temporal dynamics processing
VLSI Implementation of Cellular Neural Networks
The cellular neural nerwork (CNN) presented here is an example of very large scale analog processing or collective analog computation. The CNN architecture combines some features of fully connected analog neural networks [1,2,3] with the nearest neighbor interactions found in cellular automata [4,5,6]. A companion paper in this
SIRENA: A CAD environment for behavioural modelling and simulation of VLSI cellular neural network chips
International Journal of Circuit Theory and Applications, 1999
This paper presents SIRENA, a CAD environment for the simulation and modeling of mixed-signal VLSI parallel processing chips based on Cellular Neural Networks. SIRENA includes capabilities for: a) the description of nominal and non-ideal operation of CNN analog circuitry at the behavioral level; b) performing realistic simulations of the transient evolution of physical CNNs including deviations due to second-order effects of the hardware; and, c) evaluating sensitivity figures, and realize noise and Montecarlo simulations in the time domain.
GPU-based simulation of cellular neural networks for image processing
Neural Networks, 2009. IJCNN 2009. …, 2009
improvement translates to faster image processing algorithms compared to traditional CPU-based algorithms. topology uniform 2D grid usually feed-forward processing element dynamic equations nonlinear weighted sum common uses image processing classification, control CNNs are composed of many cells arranged in a grid, M. To simplify discussion, we will assume these grids are always square with dimensions m x m for m 2 cells. Each cell in the grid is denoted Vij for i, j EM. Thus each cell is labeled from VII to V m m . We define two types of cell depending on their location in the grid: inner cells and boundary cells. Boundary cells occur near the edges of the grid; inner cells occur everywhere else. Boundary cells necessarily have different properties than inner cells because they are connected to fewer neighboring cells. Each inner cell is the center of a neighborhood N i j of n x n cells. By this definition, n must be odd and is usually n == 3. By convention, each cell in a given neighborhood is assigned an index k from 1..n 2 , with k == 1 denoting the center cell, as shown in Figure . Thus any given center cell Vij == VI belongs Cellular neural networks (CNNs) are similar to well-known artificial neural networks (ANNs) in that they are composed of many distributed processing elements called "cells", which are connected in a network; however, there are several important differences between CNNs and ANNs (see Table ). Instead of the usual feed-forward, layered architecture seen in many types of neural networks, CNNs were designed to operate in a two-dimensional grid where each processing element (cell) is connected to neighboring cells in the grid. The cells comprising a CNN communicate by sending signals to neighboring cells in a manner similar to ANNs, but the signals are processed by each cell in a unique way. Specifically, CNN cells maintain a state which evolves through time due to differential (or difference) equations dependent on the cell's inputs and feedback. ANNs CNNs Table I CNNs, ANNs COMPARED
Cellular neural networks: Implementation of a segmentation algorithm on a Bio-inspired hardware processor
2012 IEEE 55th International Midwest Symposium on Circuits and Systems (MWSCAS), 2012
The Cellular Neural/Nonlinear Network (CNN) paradigm has recently led to a Bio-inspired (Bi-i) Cellular Vision system, which represents a computing platform consisting of sensing, array sensing-processing and digital signal processing. This paper illustrates the implementation of a novel CNN-based segmentation algorithm onto the Bi-i system. The experimental results, carried out for a benchmark video sequence, show the feasibility of the approach, which provides a frame rate of about 26 frame/sec. Finally, comparisons with existing CNN-based methods highlight that the proposed implementation represents a good trade-off between real-time requirements and accuracy.
A Hardware Coprocessor Integrated with OpenCV for Edge Detection Using Cellular Neural Networks
2011 Sixth International Conference on Image and Graphics, 2011
In this work, a high performance hardware coprocessor for CNNs and its interaction with the OpenCV library is reported. Edge detection algorithms reduce the amount of image data to be processed, because only essential information is preserved. There are several approaches for edge detection, one of them is based on Cellular Neural Networks (CNNs). The parallel nature of CNNs makes them suitable to be implemented on a reconfigurable device, such as Field Programmable Gate Arrays (FPGAs). An FPGA implementation of CNNs achieves high performance and flexibility due to fine-grain parallelism of the FPGA-based implementations. CNNs can perform both linear and nonlinear image processing tasks, such as filtering, threshold, various mathematical morphology operations, edge detection, corner detection, etc., but in this paper only the edge detection problem is addressed. Hardware resources and performance comparison are reported.
A reconfigurable architecture mapping multilayer CNN paradigms
IEEE International Workshop on Cellular Neural Networks and their Applications, 1994
A digital VLSI implementation of linear template cellular neural nets (CNNs) is presented. A reconfigurable architecture is organized as 12 layers of 64×64 cells. The CNNs are reformulated introducing sets of generalized cloning templates to enucleate more sharply the structure of both intra- and inter-layer cooperative computations. In this way it is possible to develop CNN algorithms for complex vision
Object Recognition in Image Sequences with Cellular Neural Networks
In this paper, the application of CNN associative memories for 3D object recognition is presented. The main idea is to analyse the optical flow in an image sequence of an object. Several features of the optical flow between two succeeding images are calculated and merged to a time series of features for the whole image sequence. These features show several object specific characteristics and are used for a classification step in an object recognition system. Therefore, the feature vectors of an object set are learnt and recalled by an associative memory based on the paradigm of Cellular Neural Networks, CNN.
Modified cellular neural network architecture for integrated image sensing and processing
ISCAS'99. Proceedings of the 1999 IEEE International Symposium on Circuits and Systems VLSI (Cat. No.99CH36349)
The paper describes a novel VLSI architecture adaptation of the cellular neural network (CNN) paradigm. It includes details of the design as well as test results of CMOS chip prototypes.
Unsupervised Pattern Classification by means of Cellular Neural Networks
A novel algorithm for unsupervised classification of datasets made up of integer valued patterns by means of Cellular Neural Network (CNN) is proposed. The adopted CNN is n-dimensional and is based on a space-variant template - neighborhood order 1 - to cluster n-dimensional datasets. The choice of a CNN architecture allows a straightforward hardware implementation, particularly suited for bi-dimensional patterns.
Related topics
Related papers
An Artificial Cellular Convolution Architecture for Real-Time Image Processing
ISRN Machine Vision, 2012
An artificial cell is comprised of the most basic elements in a hierarchical system, that has minimal functionality, but general enough to obey the rules of "artificial life." The ability to replicate, organize hierarchy, and generalize within an environment is some of the properties of an artificial cell. We present a hardware artificial cell having the properties of generalization ability, the ability of self-organization, and the reproducibility. The cells are used in parallel hardware architecture for implementing the realtime 2D image convolution operation. The proposed hardware design is implemented on FPGA and tested on images. We report improved processing speeds and demonstrate its usefulness in an image filtering application. O16 [0 : 31] Conv out [0 : 31] Status Fltr 3 × 3 Figure 4: The hierarchical cell architecture for 2D convolution operator.
A Scalable and Programmable Simplicial CNN Digital Pixel Processor Architecture
IEEE Transactions on Circuits and Systems I: Regular Papers, 2004
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.
Modeling multiple object scenarios for feature recognition and classification using cellular neural networks
2009
Cellular neural networks (CNNs) have been adopted in the spatio-temporal processing research field as a paradigm of complexity. This is due to the ease of designs for complex spatio-temporal tasks introduced by these networks. This has led to an increase in the adoption of CNNs for on-chip VLSI implementations. This dissertation proposes the use of a Cellular Neural Network to model, detect and classify objects appearing in multiple object scenes. The algorithm proposed is based on image scene enhancement through anisotropic diffusion; object detection and extraction through binary edge detection and boundary tracing; and object classification through genetically optimised associative networks and texture histograms. The first classification method is based on optimizing the space-invariant feedback template of the zero-input network through genetic operators, while the second method is based on computing diffusion filtered and modified histograms for object classes to generate deci...
Artificial Cellular Convolution Architecture for Real-Time Image Processing
2011
An artificial cell is comprised of the most basic elements in a hierarchical system, that has minimal functionality, but general enough to obey the rules of "artificial life." The ability to replicate, organize hierarchy, and generalize within an environment is some of the properties of an artificial cell. We present a hardware artificial cell having the properties of generalization ability, the ability of self-organization, and the reproducibility. The cells are used in parallel hardware architecture for implementing the realtime 2D image convolution operation. The proposed hardware design is implemented on FPGA and tested on images. We report improved processing speeds and demonstrate its usefulness in an image filtering application.
Image processing by three-layer cellular neural networks with a new layer arrangement
2013 IEEE International Symposium on Circuits and Systems (ISCAS2013), 2013
In this study, we research a new layer arrangement of three layer cellular neural network (CNN). In this paper, we investigate the output characteristics by using our proposed method to image processing of gray scale image and binary image and show its effectiveness with simulation results.
Cellular associative neural networks for pattern recognition
… OF YORK DEPARTMENT OF COMPUTER SCIENCE- …, 1999
A common factor of many of the problems in shape recognition and, in extension, in image interpretation is the large dimensionality of the search space. One way to overcome this situation is to partition the problem into smaller ones and combine the local solutions towards global interpretations. Using this approach, the system presented in this thesis provides a novel combination of the descriptional power of symbolic representations of image data, the parallel and distributed processing model of cellular automata and the speed and robustness of connectionist symbolic processing.
VLSI efficient discrete-time cellular neural network processor
IEE Proceedings - Circuits, Devices and Systems, 2002
Typical VLSI implementations of discrete-time cellular neural networks (DTCNN) incorporate costly hardware to implement the basic DTCNN cell, resulting in a small grid size that needs to be cascaded with many other chips for processing images of any practical size. In the paper, a low-cost DTCNN cell that can be incorporated into a single chip in large numbers has been proposed. Memory bandwidth considerations show that 256 DTCNN cells can be incorporated into a single chip DTCNN processor to compute a 256 x 256 image at 30 frames per second. Techniques based on rectangular-shaped cell grids for use with video memory have been proposed to satisfy the memory bandwidth requirements. The architecture of the proposed DTCNN processor is also capable of supporting the flexible grouping of basic cells. In addition, the processor, which is capable of supporting the flexible grouping of cells, can be cascaded in a highly scalable manner to facilitate the processing of larger images at high speed.
Cellular Neural Networks for segmentation of image sequence
Proc. the 11th Portuguese …, 2000
The Cellular Neural Networks (CNN) model is now a paradigm of cellular analog programmable multidimensional processor array with distributed local logic and memory. CNNs consist of many parallel analogue processors computing in real time. One desirable feature is that these processors arranged in a two dimensional grid only have local connections, which lend themselves easily to VLSI implementations. In this paper, we present a new algorithm for image segmentation using CNN. We start from a mathematical viewpoint (i.e., statistical regularization based on Markov Random Field, (MRF) and proceed by mapping the algorithm onto a cellular neural network. Because of the temporal dynamics inherent in the cells of the CNN it is well suited to processing time-varying images. A robust motion estimation algorithm is achieved by using a spatiotemporal neighborhood for modeling pixel interactions.
Cellular-Neural-Network Focal-Plane Processor as Pre-Processor for ConvNet Inference
2020 IEEE International Symposium on Circuits and Systems (ISCAS)
Cellular Neural Networks (CNN 1) can be embodied in the form of a focal-plane image processor. They represent a computing paradigm with evident advantages in terms of energy and resources. Their operation relies in the strong parallelization of the processing chain thanks to a distributed allocation of computing resources. In this way, image sensing and ultra-fast processing can be embedded in a single chip. This makes them good candidates for portable and/or distributed applications in fields like autonomous robots or smart cities. With the irruption of visual features learning through convolutional neural networks (ConvNets), several works attempt to implement this functionality within the CNN framework. In this paper we carry out some experiments on the implementation of ConvNets with CNN hardware in the form of a focal-plane image processor. It is shown that ultra-fast inference can be implemented, using as an example a LeNetbased ConvNet architecture.
Cellular associative neural networks for image interpretation
6th International Conference on Image Processing and its Applications, 1997
This paper describes the architecture and the operation of a neural network based system for image interpretation. The system is based on the use of two models of associative neural networks, ADAM and AURA for image and symbolic processing respectively. Employing characteristics of cellular automata theory and applying ideas from syntactic and structural pattern recognition, it uses a hierarchical approach to learn the structure of images. The hardware implementation of this system is based on the C-NNAP hardware platform.
Loading Preview
Sorry, preview is currently unavailable. You can download the paper by clicking the button above.