Efficient DTCNN implementations for large-neighborhood functions

Brugge, M.H. ter; Stevens, JH; Nijhuis, J.A.G.; Spaanenburg, Lambert

doi:10.1109/cnna.1998.685336

Cited by 6 publications

(15 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To cope with this problem, some researchers proposed to use specially-designed CNN that support large templates [29]. Others used large-template decomposition methods that use multiple smaller CNN to implement the target CNN function that requires large templates [1,8,26]. The former approach needs to design different CNNs to meet different application requirements.…”

Section: Derivation Of the Processor Core For Large-template Decomposmentioning

confidence: 99%

“…For example, a 7 × 7 (r 1 = 3) matrix is decomposed into nine 3 × 3 (r 2 = 1) matrices after zeropadding, but is decomposed into four 5 × 5 (r 2 = 2) matrices. The former (r 2 = 1) Table 1 The complexity of different template decomposition methods (r 1 = neighborhood size of the larger template, r 2 = 1 = neighborhood size of the small template) Slot [26] Brugge [1] Fernandez [8] Proposed…”

Section: Derivation Of the Processor Core For Large-template Decomposmentioning

confidence: 99%

“…Other methods also listed for comparison are Slot [26], Brugge [1], and Fernandez [8]. Among them, the method proposed by Brugge was used only for morphological processing (e.g.…”

Section: Derivation Of the Processor Core For Large-template Decomposmentioning

confidence: 99%

“…Two categories of methods were regularly employed to solve the large template problem. One is directly building large templates in the hardware [29] and the other is decomposing the large templates into smaller ones [1,8,26]. The former requires redesigning the hardware for different sizes of templates, which is usually time and manpower consuming and may be constrained by hardware resources.…”

Section: Introductionmentioning

confidence: 99%

“…Research on CNN template decomposition was probably stemmed from Slot's work in 1994 [26]. Subsequently, Brugge et al proposed another method to decompose large templates, but this method is only limited to morphological processing images [1]. Fernandez et al [8] proposed to use separate decomposition and shift templates in conjunction with the fixed and the shifted image algorithms to implement template decomposition.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

A Configurable Digital Cellular Neural Network with Template Decomposition

Lin

Hsu

2010

Circuits Syst Signal Process

View full text Add to dashboard Cite

A configurable digital cellular neural network (DCNN) with decomposable templates is presented. A template decomposition method based on physical circuit decomposition and synthesis is proposed to meet the large template requirement in CNN. The complexity of the template decomposition scheme is only O( (2n + 1)/3 2 ), which is significantly reduced when compared to the other methods in the literature. Moreover, the proposed method supports both binary and grayscale images.The silicon intellectual property (SIP) concept is employed in the design and a friendly graphical user interface (GUI) is developed to facilitate the operation. When implemented on an embedded system with Virtex II XC2V2000 chip, the DCNN is realized with 48014 gates at 18-bit data format and can run at an optimal frequency of 90 MHz. The result outperforms the other approaches to implement DCNN with much reduced gate counts and higher working frequency.

show abstract

Section: Derivation Of the Processor Core For Large-template Decomposmentioning

confidence: 99%

Section: Derivation Of the Processor Core For Large-template Decomposmentioning

confidence: 99%

“…Other methods also listed for comparison are Slot [26], Brugge [1], and Fernandez [8]. Among them, the method proposed by Brugge was used only for morphological processing (e.g.…”

Section: Derivation Of the Processor Core For Large-template Decomposmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A Configurable Digital Cellular Neural Network with Template Decomposition

Lin

Hsu

2010

Circuits Syst Signal Process

View full text Add to dashboard Cite

show abstract

Split and shift methodology on cellular processor arrays: area saving versus time penalty

Fernandez

Brea

Cabello

2012

Circuit Theory & Apps

View full text Add to dashboard Cite

This paper addresses the so-called split and shift methodology. This methodology deals with the implementation of kernels of sizes that go above the physically implemented resources (local connections and weighting circuits) on synchronous cellular processor arrays (CPA), including the realization of large neighborhood operations and/or the reduction of the available hardware in order to drop the area consumption. Two main goals are pursued in the development of the methodology, namely: (1) minimum penalty at processing time and (2) absolutely no penalty at functional level. The paper presents different techniques and guidelines for the methodology application and introduces a Figure of Merit to evaluate them by relating area gains with time penalty. This, along with a kernel shape analysis, led us to propose more adequate configurations of weighting circuits and to justify the classical choice of North-East-West-South connectivity. To validate the methodology, we realize several estimates over actual physical implementations, and we propose the realization over CPAs of the spin filters, scale invariant feature transform and speeded-up robust features algorithms. A more in-depth trade-off analysis is realized over the implementation of the pixel level snakes algorithm.but it can interact with separated PEs, thanks to the propagative effects of the array dynamics of kernel application. The basic characteristic of local connectivity and its simple SIMD control makes this kind of systems very suitable for hardware implementation. However, as a consequence, the natural size of the kernels to be applied is limited to the smallest one (3 Â 3). This is, on the other hand, an important limitation in the functionality of a CPA as larger neighborhoods are needed in several image processing primitives as diffusion or low-pass filtering operations [1], halftoning [2], texture analysis [3] or matching and hit&miss operations [4], some of them used in algorithms like modern scale-and rotation-invariant feature extractors like scale invariant feature transform (SIFT) and speeded-up robust features (SURF) [5,6].As indicated, a CPA can realize global processing taking into account the whole image information thanks to the propagative effects of the architecture. According to this we can think in solving the remote neighbors interaction through the recursive application of templates. In fact, a recursive process can be summarized in the application of a large neighborhood (LN) template. Nevertheless, the inverse process, the decomposition of a LN template into minimum-sized templates is not trivial, and different approaches have been developed to deal with this issue.The challenge is then to implement any kind of LN operations while keeping the local connectivity and with affordable penalties in performance. Our goal is to do it, in addition, with the minimum impact in the architecture at hardware level.On the other hand, on focal-plane processors with a pixel-to-PE assignment, the area occupation is not only a matter of cost or h...

show abstract

Preprocessor embedding in cellular neural network

Lin

2009

Journal of the Chinese Institute of Engineers

View full text Add to dashboard Cite

Cellular neural network (CNN) is a highly parallel analog circuit suitable for real-time image processing. However, the needs of preprocessors in real-world applications usually destroy the parallel processing nature of CNN. In this paper, we propose methods to embed the function of a preprocessor into a single CNN universal machine (CNN-UM) such that the novel CNN contains the functions of both the preprocessor and the original CNN. The embedding processes are divided into three categories according to the structure of the B template in CNN. In the general case of nonzero B template (Case 1), an unbiased method and three approximation methods are proposed. The unbiased method is equivalent to its sequential counterpart, although templates of twice the neighborhood size should be used. Among the three approximation methods, AM III shows the least deviation from the unbiased method.In some experiments, AM II and AM III even outperform the unbiased method in noise-elimination. In the two degenerated cases of Case 2 and Case 3 CNN, the embedding process can be fulfilled with templates of the original size. An exact solution can be derived for Case 2 while an approximate solution is developed for Case 3. The deviation due to the approximation in Case 3 is minor and the assumption made can even assist in noise elimination.

show abstract

Efficient DTCNN implementations for large-neighborhood functions

Cited by 6 publications

References 12 publications

A Configurable Digital Cellular Neural Network with Template Decomposition

A Configurable Digital Cellular Neural Network with Template Decomposition

Split and shift methodology on cellular processor arrays: area saving versus time penalty

Preprocessor embedding in cellular neural network

Contact Info

Product

Resources

About