A parameter representation scheme is proposed for fast and efficient Hough transform of circles. This approach replaces the conventional three-dimensional accumulator array by a pair of two-dimensional planes. Inverse...
详细信息
A parameter representation scheme is proposed for fast and efficient Hough transform of circles. This approach replaces the conventional three-dimensional accumulator array by a pair of two-dimensional planes. Inverse mapping from the estimated circle parameter sets to a subspace of the original image is used to verify the candidate circles, giving a recognition rate of 95-100%.
Two neuron growth rules are proposed: the neuron splitting rule and the neuron formation rule for self-organization of adaptive neural networks. Using these two rules the author extends the model of the neocognitron s...
详细信息
Two neuron growth rules are proposed: the neuron splitting rule and the neuron formation rule for self-organization of adaptive neural networks. Using these two rules the author extends the model of the neocognitron so that when it is presented with part of a stored pattern or with a composite image consisting of many stored patterns, the pattern will be recognized properly. The extended model does pattern partitioning when it recognizes that the input image is a part of a stored pattern, and pattern composition or association when many stored patterns are presented at the same time. This extended neocognitron is demonstrated by computer simulation. The author suggests that the proposed mechanism is a plausible model for child cognitive development and useful for visual patternrecognition.
The ALOPEX process is a broadly defined optimization procedure which simultaneously varies several parameters based on single value feedback and noise application. ALOPEX is used to solve an imagerecognition problem ...
详细信息
The ALOPEX process is a broadly defined optimization procedure which simultaneously varies several parameters based on single value feedback and noise application. ALOPEX is used to solve an imagerecognition problem using several hexagonally based templates. Efficiency improvements are examined, such as a 'temperature' effect, arrival at optimal parameter values, and a form of parallelization. Two different forms of the 'cost' function are analyzed, and a hierarchical system is attempted to attain faster 'good' solutions. The analysis is expanded to multiple template simulations, where the algorithm is able to make appropriate recognition decisions on noisy or incomplete data.
Several theorems on image transformations are proved, and new algorithms are proposed to perform these functions. These algorithms perform mapping and filling at the same time, while respecting the connectivity of the...
详细信息
ISBN:
(纸本)0818619406
Several theorems on image transformations are proved, and new algorithms are proposed to perform these functions. These algorithms perform mapping and filling at the same time, while respecting the connectivity of the original image. As a result, the transformations become more consistent and accurate. The essential parallelism in the new algorithms also facilitates their implementation using VLSI architecture, such that the time complexity is the only O(N) compared with O(N2) using a uniprocessor, where n is the dimension of the image plane. The new algorithms can handle all kinds of images, including those of long narrow objects which present problems to other algorithms. They also reduce the errors introduced by the order in which rotation and scaling are applied. A series of experiments was conducted to verify the performance of the proposed algorithms. The results indicate that the new algorithms and VLSI architectures can be very useful to imageprocessing, patternrecognition, and related areas, especially real-time applications.
Parallel matrix multiplication algorithms (based on the common data distribution formats) used in patternrecognition, image-processing, and signal-processing applications are discussed. A novel algorithm is introduce...
详细信息
Parallel matrix multiplication algorithms (based on the common data distribution formats) used in patternrecognition, image-processing, and signal-processing applications are discussed. A novel algorithm is introduced and is shown to be the fastest one for a determined class of applications. The algorithms are analyzed for performance as a function of array dimension, data distribution formats, and the architecture of the computer upon which the algorithms are executed. Performance bounds and speedups (linear in the number of processors) are established. The results of the analysis are given both as characterizations of executions on selected classes of architectures and also in the form of theorems which establish the relative performance of the algorithms across classes of data distributions and architectures.
A fast adaptive Hough transform (FAHT) approach is developed for detecting shapes which can be characterized by two parameters. This class of shapes includes both linear and circular image features. The method is base...
详细信息
A fast adaptive Hough transform (FAHT) approach is developed for detecting shapes which can be characterized by two parameters. This class of shapes includes both linear and circular image features. The method is based on identifying linear and circular segments in images by searching for clusters of evidence in two-dimensional parameter spaces. The FAHT differs from HT in the degree of freedom allowed in the placement and choice of shape of the window which defines the range of parameters under study at each resolution. This method is superior to that of HT implementation in both storage and computational requirements. The ideas of the FAHT are illustrated by tackling the problem of identifying linear segments in images by searching for clusters of evidence in two-dimensional parameter spaces. It is shown that the method is robust to the addition of extraneous noise and can be used to analyze complex images containing more than one shape.
A neural network processing scheme is proposed which utilizes a self-organizing Kohonen feature map as the front end to a feedforward classifier network. The results of a series of benchmarking studies based upon arti...
详细信息
A neural network processing scheme is proposed which utilizes a self-organizing Kohonen feature map as the front end to a feedforward classifier network. The results of a series of benchmarking studies based upon artificial statistical patternrecognition tasks indicate that the proposed architecture performs significantly better than do conventional feedforward classifier networks when the decision regions are disjoint. This is attributed to the fact that the self-organization process allows internal units in the succeeding classifier network to be sensitive to a specific set of features in the input space at the outset of training.
Summary form only given, as follows. An approach to distortion-invariant target identification and target classification based on the adaptive resonance theory II (ART-II) is presented. The neural network used demonst...
详细信息
Summary form only given, as follows. An approach to distortion-invariant target identification and target classification based on the adaptive resonance theory II (ART-II) is presented. The neural network used demonstrates fast unsupervised learning coupled with stable retention and retrieval of information. To keep the dimensions of the network to a bare minimum and to speed up the computational process as well as to achieve invariance, six distortion-invariant features are extracted from each target image and are used as network inputs. These continuous valued features are derived from the geometrical moments of the image. The ART-based target recognition system (ATRS) can be used in two different modes, a) as a target classifier and b) as a target recognition system. Several parameters associated with the network itself allow for greater flexibility in feature manipulation. The ATRS is found to perform well on three major counts: speed of processing, flexibility in feature manipulation, and noise tolerance. The determination of critical settings of the various parameters associated with the network is crucial in tuning the ATRS to ensure stable and consistent performance.
A new model which permits visual patterns to be invariant to affine transforms (translations, rotations, and dimensions) is presented. A training multilayer fully connected network of ADALINE neurons is proposed as a ...
详细信息
A new model which permits visual patterns to be invariant to affine transforms (translations, rotations, and dimensions) is presented. A training multilayer fully connected network of ADALINE neurons is proposed as a preprocessing step for invariant image extraction. A second neural network has been trained by the popular backpropagation algorithm for recovering the real image without distortions. First, the sample invariants are obtained by the preprocessing network. In the second step, the general invariant that includes all the sample invariants is computed. Afterward, the reordered sample invariants are input to a multilayer neural network trained by the backpropagation algorithm. The original image, without distortions, is obtained in the output of this system. Several test images have been computed, and evaluation of the results shows that in the case of images with intrinsic perceptual similarity, the learning procedure leads to a global invariant extraction that requires less computational effort in comparison with an arbitrary training selection. After the training process, this system is able to extract the generalized invariant image from an arbitrary picture recovering the input image without distortions.
Self-organizing maps have a connection with traditional vector quantization. A characteristic which makes them resemble certain biological brain maps, however, is the spatial order of their responses which is formed i...
详细信息
Self-organizing maps have a connection with traditional vector quantization. A characteristic which makes them resemble certain biological brain maps, however, is the spatial order of their responses which is formed in the learning process. Two innovations are discussed: dynamic weighting of the input signals at each input of each cell, which improves the ordering when very different input signals are used, and definition of neighborhoods in the learning algorithm by the minimum spanning tree, which provides a far better and faster approximation of prominently structured density functions. It is cautioned that if the maps are used for patternrecognition and decision processes, it is necessary to fine-tune the reference vectors such that they directly define the decision borders.
暂无评论