Scale Invariant Pattern Recognition Research Articles

Robust Partial Shape Recognition Using Curvature and an Iterative Closest Point Algorithm

This article presents a robust translation, rotation, and scaling invariant pattern recognition algorithm using partially distorted two-dimensional (2D) contours. Both global and local matching accuracies are guaranteed: the former is achieved by using global control point (GCP) and curvature matching and the latter by using the iterative closest point (ICP) algorithm. In the proposed algorithm, an input contour is considered two possible types: first the contour contains GCP and the second contains no GCP. For the contours with no detected GCPs, the algorithm considers the contour as the second type of contour and the curvature matching algorithm is directly applied. For the contour with GCPs, a global alignment is performed using the GCPs and then, the curvature matching algorithm is applied. The validity of the algorithm is illustrated by the presentation of experimental results. The experiments show that the algorithm works well for partial object recognition with and without GCPs.

Shift and scale‐invariant correlation for pattern recognition based on the bidimensional empirical mode decomposition with noise robustness

A novel filter is proposed to improve the noise robustness and discrimination capability for shift and scale invariant pattern recognition. This filter is a combination of mellin radial harmonic filter (RHF) and the bidimensional empirical mode decomposition. The basic principle of this method is to make use of partial reconstructions of the image by the relevant intrinsic mode functions corresponding to the most important structures of the image. A criterion is proposed to determine the proper number of intrinsic mode functions to be discarded for denoising by discussing the characteristic of the noise. The proposed filter provides a wider allowable scale change of the object. Within this range, the correlation peak intensity is relatively uniform even in the case of noise. This proposed filter has been tested experimentally to confirm the result from numerical simulations for cases with and without additive white Gaussian noise.

Shift- and scale-invariant pattern recognition using morphological fringe-adjusted joint transform correlation

A shift- and scale-invariant version of morphological fringe-adjusted joint transform correlation (MFJTC) is proposed in this paper, which we call shift- and scale-invariant MFJTC (SIMFJTC). SIMFJTC is combination of MFJTC and conventional radial harmonic filter (RHF). Using computer simulation, we compare the output results of SIMFJTC with those of morphological radial harmonic correlation (MRHC) and SDF-based FPFJTC when input scene is corrupted by salt-and-pepper noise and cluttered background. Our results show that SIMFJTC has higher discriminability and displays better stability against salt-and-pepper noise and cluttered background. Moreover, scale-invariance of SIMFJTC is much stricter than MRHC and SDF-based FPFJTC.

Scale-invariant pattern recognition using a combined Mellin radial harmonic function and the bidimensional empirical mode decomposition

A novel scale and shift invariant pattern recognition method is proposed to improve the discrimination capability and noise robustness by combining the bidimensional empirical mode decomposition with the Mellin radial harmonic decomposition. The flatness of its peak intensity response versus scale change is improved. This property is important, since we can detect a large range of scaled patterns (from 0.2 to 1) using a global threshold. Within this range, the correlation peak intensity is relatively uniform with a variance below 20%. This proposed filter has been tested experimentally to confirm the result from numerical simulation for cases both with and without input white noise.

Scale invariant recognition in optical correlators using prior estimation of object size

A new perspective of scale-invariant pattern recognition is proposed in this paper. Instead of computing a filter from one of many orthogonal components of the reference object, the scaling factor is estimated from the Fourier spectrums of reference and test images and utilized in rescaling of the spectrum. The test image is correlated with the reference image after rescaling its Fourier spectrum with classical matched filtering. Computer simulation results are presented for a reference object at several scaling versions as well as some non-target objects.

Improved performance of scale-invariant pattern recognition using a combined Mellin radial harmonic function and wavelet transform

By using a combination of the Mellin radial harmonic function and the Mexican-hat wavelet transform, scale- and shift-invariant pattern recognition is reported. Without preprocessing of the input object, this filter is capable of identifying over a wide allowable scale range of 0.25 to 1. The correlation peaks are sharp, and the peak intensity is fairly uniform, with a variance below 30%. Computer simulation is adopted to investigate the performance of the filter. Experimental results implemented using a photorefractive joint-transform correlator, including those obtained under white noise, are also presented.

Binary phase only filters for rotation and scale invariant pattern recognition with the joint transform correlator

The joint transform correlator (JTC) is one of the two main optical image processing architecture which provides a highly effective way of comparing images in a wide range of applications. Traditionally, an optical correlator is used to compare an unknown input scene with a pre-captured reference image library, to detect if the reference occurs within the input. Strength of the correlation signal decreases rapidly as the input object rotates or varies in scale relative to the reference object. The aim of this paper is to overcome the intolerance of the JTC to rotation and scale changes in the target image. Many JTC systems are constructed with the use of ferroelectric liquid crystal (FLC) spatial light modulators (SLMs) as they provide fast two-dimensional binary modulation of coherent light. Due to the binary nature of the FLC SLMs used in the JTC systems, any image addressed to the device need to have some form of thresholding. Carefully thresholding the grey scale input plane and the joint power spectrum (JPS) has significant effect on the quality of correlation peaks and zero order (DC) noise. A new thresholding technique to binarise the JPS has been developed and implemented optically. This algorithm selectively enhances the desirable fringes in the JPS which provide correlation peaks of higher intensity. Zero order noise is further reduced when compared to existing thresholding techniques. Keeping in mind the architecture of the JTC and limitations of FLC SLMs, a new technique to design rotation and scale invariant binary phase only filters for the JTC architecture is presented. Filers design with this technique have limited dynamic range, higher discriminability among target and non-target objects, and convenience for implementation on FLC SLMs. Simulation and experiments shows excellent results of various rotation and scale invariant filters designed with this technique. A rotation invariant filter is needed for various machine vision applications of the JTC. By fixing the distance between camera and input object, the scale sensitivity of the correlator can be avoided. In contrast to the industrial machine vision applications, scale factor is very important factor for the applications of the JTC systems in defence and security. A security system using a scale invariant JTC will be able to detect a target object well in advance and will provide more time to take a decision.

Wire length as a circuit complexity measure

We introduce wire length as a salient complexity measure for analyzing the circuit complexity of sensory processing in biological neural systems. This new complexity measure is applied in this paper to two basic computational problems that arise in translation- and scale-invariant pattern recognition, and hence appear to be useful as benchmark problems for sensory processing. We present new circuit design strategies for these benchmark problems that can be implemented within realistic complexity bounds, in particular with linear or almost linear wire length. Finally, we derive some general bounds which provide information about the relationship between new complexity measure wire length and traditional circuit complexity measures.

Nonlinear radial-harmonic correlation using binary decomposition for scale-invariant pattern recognition

We introduce a new scale-invariant pattern-recognition method that uses nonlinear correlation. We applied several common linear correlations to images decomposed into disjoint binary images, which is very discriminant even when the target is embedded in strong noise. We combine our sliced orthogonal nonlinear generalized correlation method and the radial-harmonic expansion in order to achieve scale-invariant pattern recognition. The information from a radial harmonic for each binary slice of the reference object is combined with binary slices of the target. The method avoids the time-consuming process of finding expansion centers for the radial harmonics. The stability of the correlation peaks in the presence of background noise and of overlapping Gaussian noise is also studied.

Translation and scale-invariant pattern recognition based on the Hamming network

The purpose of this paper is to develop a position and scale-invariant pattern recognition system based on the Hamming network (HN). The input image is processed by a feature extraction network to calculate the exact position and dimension of the pattern feature. An associate memory network is the designed to map the obtained dimension information to its corresponding HN weight matrix. With its weight matrix determined, the HN maps the input image into a proper class. Since all network parameters are obtained without learning, the system can be realised very easily.

Morphological radial-harmonic correlation for shift- and scale-invariant pattern recognition

A novel approach is proposed to improve the discrimination of the radial-harmonic filter for shift- and scale-invariant pattern recognition. The approach combines the morphological correlation and the radial- harmonic expansion, in which the reference gray-level image is first decomposed into a series of binary images, and then the radial-harmonic expansion is applied to those binary images. The morphological radial- harmonic correlation is achieved by correlating the single-order radial- harmonic components with the corresponding sliced binary images obtained from the input scene and summing the correlations. The algorithm is tested via computer simulations. The results show that the algorithm is shift- and scale-invariant and provides better discrimination ability than the conventional radial-harmonic filter.

A hybrid learning network for shift, orientation, and scaling invariant pattern recognition

A three-layer neural network is presented as a generic approach for visual pattern recognition invariant with respect to the geometric appearance such as translation, orientation and scale of the patterns. The invariant recognition is achieved by representing the geometric variations internally in the network by nodes in the input and middle layers, which are laterally connected and trained by a hybrid algorithm combining both competitive and Hebbian learning. As the result of the hybrid learning, each pattern will be represented by a particular subset of middle-layer nodes all specialized to respond to the same pattern but with different geometric appearances. The nodes in the output layer are then trained by competitive learning to recognize the different pattern internally represented by the middle-layer nodes, independent of their location, orientation and size. The proposed algorithm is generic and robust and can be applied to various practical recognition problems. Moreover, the network is relatively simple and biologically plausible and can serve as a computational model to account for the invariant object recognition in the biological visual system.

A hybrid learning network for shift, orientation, and scaling invariant pattern recognition

A three-layer neural network is presented as a generic approach for visual pattern recognition invariant with respect to the geometric appearance such as translation, orientation and scale of the patterns. The invariant recognition is achieved by representing the geometric variations internally in the network by nodes in the input and middle layers, which are laterally connected and trained by a hybrid algorithm combining both competitive and Hebbian learning. As the result of the hybrid learning, each pattern will be represented by a particular subset of middle-layer nodes all specialized to respond to the same pattern but with different geometric appearances. The nodes in the output layer are then trained by competitive learning to recognize the different pattern internally represented by the middle-layer nodes, independent of their location, orientation and size. The proposed algorithm is generic and robust and can be applied to various practical recognition problems. Moreover, the network is relatively simple and biologically plausible and can serve as a computational model to account for the invariant object recognition in the biological visual system.

A hybrid learning network for shift, orientation, and scaling invariant pattern recognition

A three-layer neural network is presented as a generic approach for visual pattern recognition invariant with respect to the geometric appearance such as translation, orientation and scale of the patterns. The invariant recognition is achieved by representing the geometric variations internally in the network by nodes in the input and middle layers, which are laterally connected and trained by a hybrid algorithm combining both competitive and Hebbian learning. As the result of the hybrid learning, each pattern will be represented by a particular subset of middle-layer nodes all specialized to respond to the same pattern but with different geometric appearances. The nodes in the output layer are then trained by competitive learning to recognize the different pattern internally represented by the middle-layer nodes, independent of their location, orientation and size. The proposed algorithm is generic and robust and can be applied to various practical recognition problems. Moreover, the network is relatively simple and biologically plausible and can serve as a computational model to account for the invariant object recognition in the biological visual system.

On local time–frequency features of speech and their employment in speaker verification

Commonly used robust speaker verification systems 1 1 We are only concerned about text-dependent verification with cooperative speakers in a low-noise environment. are based on time-varying autoregressive spectral estimation (AR) combined with hidden Markov modeling (HMM) or dynamic time warping (DTW). An exhaustive optimization of these methods in the past has culminated in quite reliable verification schemes. It seems unlikely, though, that further significant improvements are readily obtained along the same path. While short-time AR-modeling focuses on the time-varying spectral envelope of an utterance, we are introducing a new method that focuses on high-resolution estimates of the time-varying spectral structure of individual pitch periods. The new method employs reduced interference time–frequency distributions (RIDs) in combination with a scale and translation invariant pattern recognition technique (STIR). The new method by itself does not deliver better results than commonly used techniques; however, it is shown that an acceptance/rejection decision derived from both AR-DTW and RID–STIR features greatly improves the performance of the verification system.

Scale-invariant analysis of wear particle surface morphology: III: Pattern recognition

A new scale-invariant pattern recognition method has been developed for the characterization of wear particle surface morphology. The method is based on a partition-iterated function system (PIFS) constructed for a wear particle image. A database containing PIFSs constructed for different classes of wear particles was built. Each class is a set of wear particles exhibiting a similar morphology. The objective of pattern recognition is to assign a new wear particle, which is not yet classified, into one of these particle classes. For the classification, PIFSs taken from the database are decoded with the unclassified particle image used as an initial image. Decoded images obtained from PIFSs are then individually compared with the new particle image. If the difference calculated between these images is less than a threshold value, the new particle is assigned to a class. Problems associated with the use of the pattern recognition method, such as rotational- and translation-invariance, comparison between two images and computer implementation, are addressed in this paper. The effects of image resolution, tilt angle, noise, gain variations and defocusing were also investigated. Results obtained for computer-generated images of fractal surfaces and microscopic images of wear particles are demonstrated.

Scale invariant amplitude modulated phase-only filtering

A new synthetic discriminant-function-based amplitude modulated phase-only filtering technique is proposed for scale invariant pattern recognition. This technique has been found to yield improved correlation output when compared with the alternate techniques. The proposed technique is inherently suitable for optical implementation using the currently available spatial light modulators. Simulation results confirm the effectiveness of the proposed technique.

Amplitude-modulated radial harmonic filters

An amplitude-modulated radial harmonic filter is introduced to improve the discrimination of the radial harmonic filter for shift- and scale-invariant pattern recognition. The improvement of discrimination is achieved by Fourier-domain amplitude modulation, which places more emphasis on the high-frequency terms. The gain of better discrimination, however, is at the cost of poorer noise robustness since the amplitude- modulated radial harmonic filter is an optimal trade-off filter for noise robustness and discrimination. To compensate for the loss of noise robustness, we apply the technique of Fourier-domain amplitude truncation to the proposed filter. With a given size of bandpass, the amplitude truncation approach results in the lowest signal energy loss, thus yielding a better SNR and discrimination than with a low-pass amplitude-modulated radial harmonic filter.

Extended scale-invariant pattern recognition with white-light illumination

A previous method of obtaining scale-invariance detection with white-light illumination has been improved on. We were able to detect different scaled versions of the target up to a magnification factor equal to 2. We simultaneously detected several versions in the same scene, because each scale factor is codified in a different wavelength. Experimental results demonstrate the proposed technique and show the utility of the method.

Improvement of the generalization capability for a pattern-recognition neural network that uses a Gaussian-synapse neuron model

A method for improving the generalization capability for optical pattern recognition by use of a Gaussian-synapse neuron model is discussed. By the dispersive effect of the Gaussian function the input images are blurred and then fed into a multilayer neural network for learning and recognition. The effectiveness of this method is demonstrated in two-dimensional shift- and scale-invariant optical pattern recognition.