Henri Arsenault - Academia.edu (original) (raw)

Papers by Henri Arsenault

Optical Society of America Annual Meeting, 1991

A circular-harmonic (CH) filter contains only one component of the object that gives rotation in ... more A circular-harmonic (CH) filter contains only one component of the object that gives rotation in variance. But an infinity of objects could have the same center correlation value although most of those are not realistic objects. On the other hand, a CH filter performs the CH expansion of the input about every point in the plane (r,θ). The correlation output can be expressed as where F m (p) and G m (p) are the CH functions of the filter and of the input in the Fourier plane respectively, Jm(2πpr) is the Bessel function. This equation shows clearly that apart from the center r = 0 θ = 0, the 2-D correlation of a CH filter of order m depends on the CH components of all the orders and is then unique of a given input. The center output intensity depends only on the order m. But the position of the output peak depends on all the orders.

Applied optics, 1983

... Ill. Discrimination Enhancement According to Paek and Lee, using a pseudoobject can decrease ... more ... Ill. Discrimination Enhancement According to Paek and Lee, using a pseudoobject can decrease a cross-correlation to zero under the hypothesis that gl and g2 are not overlapping. ... 1 (a) with p = 1 is marked with a cross. Shown in Figs. ...

Applied optics, Sep 15, 1986

A new filter with limited rotation invariance for real target functions is proposed. The filter i... more A new filter with limited rotation invariance for real target functions is proposed. The filter is the Fourier transform of the real part of a circular harmonic component of a target. The range of limited rotation invariance depends on the harmonic order. The lower the order, the wider the ...

Annual Meeting Optical Society of America, 1988

The scale-invariant and rotation-invariant Fourier-Mellin transform depends on the position of th... more The scale-invariant and rotation-invariant Fourier-Mellin transform depends on the position of the object. This limits its field of use. We propose filters of the form rs–2 exp(jmθ) with complex-valued s = v + jw. The filter contains no object information. The correlation function of the filter with an input object at every point (x', y') in the output plane is the Fourier-Mellin transform of the object developed about the origin (x', y') of a polar coordinate system. The shift-invariant features are then extracted: these are the distances between the correlation maxima from the filters with different orders s and m and the intensities of those maxima. The logpolar coordinate transform normally used for the optical Fourier-Mellin transform is not required. The method is invariant under changes of position, scale, orientation, rotation, and intensity. It also allows input containing multiple objects simultaneously. The invariant features can be used for image classification or as inputs to neural networks for invariant associative and adaptive pattern recognition.

Annual Meeting Optical Society of America, 1988

A circular harmonic (CH) filter invariant to changes of position and orientation is described in ... more A circular harmonic (CH) filter invariant to changes of position and orientation is described in polar coordinates (ß,ψ) as Fm(ß) exp(jmψ), where Fm(ß) is the Hankel transform of the CH functions of the object. For phase-only CH filters, the function exp(jmψ) ensures the rotation invariance, and the phase of Fm(ß) contains the object information. A circular disk on the Fourier plane covers the CH filter in those segments of ß where |Fm(ß)| is small. This can improve the SNR and the discrimination ability. Experimental studies on the effect of input noise, background, and binarization of the phase-only CH filters are presented.

Nucleation and Atmospheric Aerosols, 2006

ABSTRACT We present a novel method of three-dimensional object recognition independent of illumin... more ABSTRACT We present a novel method of three-dimensional object recognition independent of illumination conditions. The recognition model is based on a vector space representation using an orthonormal basis generated by the Lambertian reflectance functions obtained with distant light sources. The recognition method proposed is based on the calculation of the angle between the vector associated with a certain illuminated three-dimensional object and a vector subspace. We define the angle in terms of linear correlations in order to get shift and illumination-invariant detection.

Optics Communications, Jun 1, 1986

ABSTRACT

Applied optics, Jul 1, 1975

Previous studies have shown-that the intensity of the correlation peaks in Vander Lugt optical co... more Previous studies have shown-that the intensity of the correlation peaks in Vander Lugt optical correlators can change with the position of the object because of emulsion thickness effects. In this paper, it is shown that the intensity variations are caused by a combination of thickness effects and a position-dependent frequency cutoff due to space variance of the optical processor. Experiments showing the global effect, as well as experiments showing each effect acting alone, have been performed.

Optical Engineering, Dec 1, 1984

Matched filters with signal-to-noise ratios that are space-invariant and rotation-invariant with ... more Matched filters with signal-to-noise ratios that are space-invariant and rotation-invariant with respect to the target have been developed. Our approach has been to extract from the target one or more circular harmonic components and to use a filter matched to these components. The technique has been implemented both digitally and with an optical processor using com-puter-generated holograms. To discriminate between objects that are almost similar, a pseudo-object is used to generate the matched filter. The technique was introduced previously as an ad hoc method; a more systematic approach based on system constraints suggests the existence of other rotation-invariant linear filters.

Applied optics, Mar 15, 1984

Multiple circular harmonic components of the same target can be used jointly to discriminate betw... more Multiple circular harmonic components of the same target can be used jointly to discriminate between the target and other objects. Two methods are considered: Coherent superposition and decision making in multidimensional space. Experimental results for the discrimination between a tank and a truck are given.

Journal of the Optical Society of America, Apr 1, 1975

Journal of the Optical Society of America, 1983

Abstract A phase reconstruction method based on the two-dimensional sampling theorem is applied t... more Abstract A phase reconstruction method based on the two-dimensional sampling theorem is applied to the case when only the modulus of the Fourier transform and the support of the object are known. The convengence of the algorithm is analysed for different intermediate sampling grids and for different initial values.

Annual Meeting Optical Society of America, 1987

A rotation-invariant single circular harmonic filter does not in general yield a maximum center a... more A rotation-invariant single circular harmonic filter does not in general yield a maximum center autocorrelation peak. A new method for predicting the level and position of the center correlation and of sidelobes as a function of the expansion center of the circular harmonic filter was recently developed.1 The proper expansion center that ensures the maximum peak correlation can be easily found by this method. The position of the correlation peak corresponds to the expansion center for the maximum energy of the circular harmonic component. The positions of the correlation peaks of different circular harmonic filters constitute a feature pattern of the object that can be used for rotation-invariant detection of the object against sidelobes yielded by noisy background and parasite objects in a complex scene. The discrimination ability is much improved by the use of multiple circular harmonic components. Only intensity thresholding, clustering, and matching of the simple point patterns are needed for the output of the filters. The procedure is faster and easier than the other methods.

Optical Computing

Three-dimensional optical multistage interconnection networks (MINs) can dynamically connect a 2-... more Three-dimensional optical multistage interconnection networks (MINs) can dynamically connect a 2-D array of NxN inputs to a 2-D array of NxN outputs. They exploit the 3-D nature of light propagation in free space for high density and high speed interconnections which are difficult to implement with planar electronic VLSI technology.

Annual Meeting Optical Society of America

A new three-layer neural network is proposed. A special structure of connections between the laye... more A new three-layer neural network is proposed. A special structure of connections between the layer of input nodes and the intermediate layer forms the memory matrix of the network. The intermediate layer is composed of neuronlike threshold elements and is totally interconnected by feedback loops. The weights of the lateral inhibition connections are determined by the cross correlations of the stored vectors. Programming the interconnections between the intermediate layer and an output slab allows any arbitrary output for any input. The network was designed to perform as a content-addressable memory or, for example, as a symbolic substitution system with a winner-take-all function, which selects the neuron with the largest activity. Such an architecture has been shown1,2 to be a better neural network processor than the widely known and discussed Hopfield model.

Annual Meeting Optical Society of America, 1988

A new rotationally invariant matched filter has been developed. The principle is based on the ext... more A new rotationally invariant matched filter has been developed. The principle is based on the extraction from the target of the information based on rotation-invariant principal components. For this purpose, the Karhunen-Loeve expansion applied to the target f(r, θ) sampled on angular radii yields its representation with a set of radial eigenfunctions ordered by decreasing eigenvalues: where Ψ k (θ) denotes the kth eigenvector of the angular covariance matrix. The inverse Karhunen-Loeve transform using a reduced number of radial eigenfunctions yields the synthesis of a new pseudoobject defined by The similarity between the original and the reconstructed targets depends on both the order and the number of the radial eigenfunctions used in the inverse transform. A rotationally invariant filter is then obtained by performing the circular harmonic expansion of the reconstructed pseudoobject.

Optica Pura y Aplicada, 1997

We use an optoelectronic pattern recognition system to combine multiple image informationonto a s... more We use an optoelectronic pattern recognition system to combine multiple image informationonto a single binary hologram, by compressing the 2D images into near-lD vectors using a special ring target. These vectors are fed 1n to an optical neural network trained with a learning rule tbat minimizes the lateral activations that would that Would interfere with recognition. We present optical results showing the usefulness of this technique.

Optics Communications, 1996

A comparison of the performance of Locally Nonlinear Filters (LNMFs) and Optimal Tradeoff Filters... more A comparison of the performance of Locally Nonlinear Filters (LNMFs) and Optimal Tradeoff Filters (OTFs) for the recognition of single images has been carried out. The comparison is done using several performance criteria such as Signal to Noise Ratio, Peak to Correlation Energy Ratio and Optical Efficiency. A strong si~Ia~ty between the whitening strategies of both filter families has been found wheu the Optimal Tmdeo~Filters for peak sharpness and noise robustness are compared with LNMFs. Our tests show a marginal superiority of OTFs over LNh4Fs.

Optical Society of America Annual Meeting, 1991

A circular-harmonic (CH) filter contains only one component of the object that gives rotation in ... more A circular-harmonic (CH) filter contains only one component of the object that gives rotation in variance. But an infinity of objects could have the same center correlation value although most of those are not realistic objects. On the other hand, a CH filter performs the CH expansion of the input about every point in the plane (r,θ). The correlation output can be expressed as where F m (p) and G m (p) are the CH functions of the filter and of the input in the Fourier plane respectively, Jm(2πpr) is the Bessel function. This equation shows clearly that apart from the center r = 0 θ = 0, the 2-D correlation of a CH filter of order m depends on the CH components of all the orders and is then unique of a given input. The center output intensity depends only on the order m. But the position of the output peak depends on all the orders.

Applied optics, 1983

... Ill. Discrimination Enhancement According to Paek and Lee, using a pseudoobject can decrease ... more ... Ill. Discrimination Enhancement According to Paek and Lee, using a pseudoobject can decrease a cross-correlation to zero under the hypothesis that gl and g2 are not overlapping. ... 1 (a) with p = 1 is marked with a cross. Shown in Figs. ...

Applied optics, Sep 15, 1986

A new filter with limited rotation invariance for real target functions is proposed. The filter i... more A new filter with limited rotation invariance for real target functions is proposed. The filter is the Fourier transform of the real part of a circular harmonic component of a target. The range of limited rotation invariance depends on the harmonic order. The lower the order, the wider the ...

Annual Meeting Optical Society of America, 1988

The scale-invariant and rotation-invariant Fourier-Mellin transform depends on the position of th... more The scale-invariant and rotation-invariant Fourier-Mellin transform depends on the position of the object. This limits its field of use. We propose filters of the form rs–2 exp(jmθ) with complex-valued s = v + jw. The filter contains no object information. The correlation function of the filter with an input object at every point (x', y') in the output plane is the Fourier-Mellin transform of the object developed about the origin (x', y') of a polar coordinate system. The shift-invariant features are then extracted: these are the distances between the correlation maxima from the filters with different orders s and m and the intensities of those maxima. The logpolar coordinate transform normally used for the optical Fourier-Mellin transform is not required. The method is invariant under changes of position, scale, orientation, rotation, and intensity. It also allows input containing multiple objects simultaneously. The invariant features can be used for image classification or as inputs to neural networks for invariant associative and adaptive pattern recognition.

Annual Meeting Optical Society of America, 1988

A circular harmonic (CH) filter invariant to changes of position and orientation is described in ... more A circular harmonic (CH) filter invariant to changes of position and orientation is described in polar coordinates (ß,ψ) as Fm(ß) exp(jmψ), where Fm(ß) is the Hankel transform of the CH functions of the object. For phase-only CH filters, the function exp(jmψ) ensures the rotation invariance, and the phase of Fm(ß) contains the object information. A circular disk on the Fourier plane covers the CH filter in those segments of ß where |Fm(ß)| is small. This can improve the SNR and the discrimination ability. Experimental studies on the effect of input noise, background, and binarization of the phase-only CH filters are presented.

Nucleation and Atmospheric Aerosols, 2006

ABSTRACT We present a novel method of three-dimensional object recognition independent of illumin... more ABSTRACT We present a novel method of three-dimensional object recognition independent of illumination conditions. The recognition model is based on a vector space representation using an orthonormal basis generated by the Lambertian reflectance functions obtained with distant light sources. The recognition method proposed is based on the calculation of the angle between the vector associated with a certain illuminated three-dimensional object and a vector subspace. We define the angle in terms of linear correlations in order to get shift and illumination-invariant detection.

Optics Communications, Jun 1, 1986

ABSTRACT

Applied optics, Jul 1, 1975

Previous studies have shown-that the intensity of the correlation peaks in Vander Lugt optical co... more Previous studies have shown-that the intensity of the correlation peaks in Vander Lugt optical correlators can change with the position of the object because of emulsion thickness effects. In this paper, it is shown that the intensity variations are caused by a combination of thickness effects and a position-dependent frequency cutoff due to space variance of the optical processor. Experiments showing the global effect, as well as experiments showing each effect acting alone, have been performed.

Optical Engineering, Dec 1, 1984

Matched filters with signal-to-noise ratios that are space-invariant and rotation-invariant with ... more Matched filters with signal-to-noise ratios that are space-invariant and rotation-invariant with respect to the target have been developed. Our approach has been to extract from the target one or more circular harmonic components and to use a filter matched to these components. The technique has been implemented both digitally and with an optical processor using com-puter-generated holograms. To discriminate between objects that are almost similar, a pseudo-object is used to generate the matched filter. The technique was introduced previously as an ad hoc method; a more systematic approach based on system constraints suggests the existence of other rotation-invariant linear filters.

Applied optics, Mar 15, 1984

Multiple circular harmonic components of the same target can be used jointly to discriminate betw... more Multiple circular harmonic components of the same target can be used jointly to discriminate between the target and other objects. Two methods are considered: Coherent superposition and decision making in multidimensional space. Experimental results for the discrimination between a tank and a truck are given.

Journal of the Optical Society of America, Apr 1, 1975

Journal of the Optical Society of America, 1983

Abstract A phase reconstruction method based on the two-dimensional sampling theorem is applied t... more Abstract A phase reconstruction method based on the two-dimensional sampling theorem is applied to the case when only the modulus of the Fourier transform and the support of the object are known. The convengence of the algorithm is analysed for different intermediate sampling grids and for different initial values.

Annual Meeting Optical Society of America, 1987

A rotation-invariant single circular harmonic filter does not in general yield a maximum center a... more A rotation-invariant single circular harmonic filter does not in general yield a maximum center autocorrelation peak. A new method for predicting the level and position of the center correlation and of sidelobes as a function of the expansion center of the circular harmonic filter was recently developed.1 The proper expansion center that ensures the maximum peak correlation can be easily found by this method. The position of the correlation peak corresponds to the expansion center for the maximum energy of the circular harmonic component. The positions of the correlation peaks of different circular harmonic filters constitute a feature pattern of the object that can be used for rotation-invariant detection of the object against sidelobes yielded by noisy background and parasite objects in a complex scene. The discrimination ability is much improved by the use of multiple circular harmonic components. Only intensity thresholding, clustering, and matching of the simple point patterns are needed for the output of the filters. The procedure is faster and easier than the other methods.

Optical Computing

Three-dimensional optical multistage interconnection networks (MINs) can dynamically connect a 2-... more Three-dimensional optical multistage interconnection networks (MINs) can dynamically connect a 2-D array of NxN inputs to a 2-D array of NxN outputs. They exploit the 3-D nature of light propagation in free space for high density and high speed interconnections which are difficult to implement with planar electronic VLSI technology.

Annual Meeting Optical Society of America

A new three-layer neural network is proposed. A special structure of connections between the laye... more A new three-layer neural network is proposed. A special structure of connections between the layer of input nodes and the intermediate layer forms the memory matrix of the network. The intermediate layer is composed of neuronlike threshold elements and is totally interconnected by feedback loops. The weights of the lateral inhibition connections are determined by the cross correlations of the stored vectors. Programming the interconnections between the intermediate layer and an output slab allows any arbitrary output for any input. The network was designed to perform as a content-addressable memory or, for example, as a symbolic substitution system with a winner-take-all function, which selects the neuron with the largest activity. Such an architecture has been shown1,2 to be a better neural network processor than the widely known and discussed Hopfield model.

Annual Meeting Optical Society of America, 1988

A new rotationally invariant matched filter has been developed. The principle is based on the ext... more A new rotationally invariant matched filter has been developed. The principle is based on the extraction from the target of the information based on rotation-invariant principal components. For this purpose, the Karhunen-Loeve expansion applied to the target f(r, θ) sampled on angular radii yields its representation with a set of radial eigenfunctions ordered by decreasing eigenvalues: where Ψ k (θ) denotes the kth eigenvector of the angular covariance matrix. The inverse Karhunen-Loeve transform using a reduced number of radial eigenfunctions yields the synthesis of a new pseudoobject defined by The similarity between the original and the reconstructed targets depends on both the order and the number of the radial eigenfunctions used in the inverse transform. A rotationally invariant filter is then obtained by performing the circular harmonic expansion of the reconstructed pseudoobject.

Optica Pura y Aplicada, 1997

We use an optoelectronic pattern recognition system to combine multiple image informationonto a s... more We use an optoelectronic pattern recognition system to combine multiple image informationonto a single binary hologram, by compressing the 2D images into near-lD vectors using a special ring target. These vectors are fed 1n to an optical neural network trained with a learning rule tbat minimizes the lateral activations that would that Would interfere with recognition. We present optical results showing the usefulness of this technique.

Optics Communications, 1996

A comparison of the performance of Locally Nonlinear Filters (LNMFs) and Optimal Tradeoff Filters... more A comparison of the performance of Locally Nonlinear Filters (LNMFs) and Optimal Tradeoff Filters (OTFs) for the recognition of single images has been carried out. The comparison is done using several performance criteria such as Signal to Noise Ratio, Peak to Correlation Energy Ratio and Optical Efficiency. A strong si~Ia~ty between the whitening strategies of both filter families has been found wheu the Optimal Tmdeo~Filters for peak sharpness and noise robustness are compared with LNMFs. Our tests show a marginal superiority of OTFs over LNh4Fs.