Max B. Reid

Coarse-coded higher-order neural networks for PSRI object recognition

The authors describe a coarse coding technique and present simulation results illustrating its usefulness and its limitations. Simulations show that a third-order neural network can be trained to distinguish between two objects in a 4096x4096 pixel input field independent of transformations in translation, in-plane rotation, and scale in less than ten passes through the training set. Furthermore, the authors empirically determine the limits of the coarse coding technique in the position, scale, and rotation invariant (PSRI) object recognition domain.

Robust position, scale, and rotation invariant object recognition using higher-order neural networks

Abstract For object recognition invariant to changes in the objects position, size, and in-plane rotation, higher-order neural networks (HONNs) have numerous advantages over other neural network approaches. Because distortion invariance can be built into the architecture of the network, HONNs need to be trained on just one view of each object, not numerous distorted views, reducing the training time significantly. Further, 100% accuracy can be guaranteed for noise-free test images characterized by the built-in distortions. Specifically, a third-order neural network trained on just one view of an SR-71 aircraft and a U-2 aircraft in a 127 × 127 pixel input field successfully recognized all views of both aircraft larger than 70% of the original size, regardless of orientation or position of the test image. Training required just six passes. In contrast, other neural network approaches require thousands of passes through a training set consisting of a much larger number of training images and typically achieve only 80–90% accuracy on novel views of the objects. The above results assume a noise-free environment. The performance of HONNs is explored with non-ideal test images characterized by white Gaussian noise or partial occlusion. With white noise added to images with an ideal separation of background vs. foreground gray levels, it is shown that HONNs achieve 100% recognition accuracy for the test set for a standard deviation up to ∼10% of the maximum gray value and continue to show good performance (defined as better than 75% accuracy) up to a standard deviation of ∼14%. HONNs are also robust with respect to partial occlusion. For the test set of training images with very similar profiles, HONNs achieve 100% recognition accuracy for one occlusion of ∼13% of the input field size and four occlusions of ∼70% of the input field size. They show good performance for one occlusion of ∼23% of the input field size or four occlusions of ∼15% of the input field size each. For training images with very different profiles, HONNs achieve 100% recognition accuracy for the test set for up to four occlusions of ∼2% of the input field size and continue to show good performance for up to four occlusions of ∼23% of the input field size each.

Connectivity strategies for higher-order neural networks applied to pattern recognition

Different strategies for non-fully connected HONNs (higher-order neural networks) are discussed, showing that by using such strategies an input field of 128×128 pixels can be attained while still achieving in-plane rotation and translation-invariant recognition. These techniques allow HONNs to be used with the larger input scenes required for practical pattern-recognition applications. The number of interconnections that must be stored has been reduced by a factor of approximately 200000 in a T/C case and ~2000 in a Space Shuttle/F-18 case by using regional connectivity. Third-order networks have been simulated using several connection strategies

Higher-Order Neural Networks Applied to 2D and 3D Object Recognition

A higher-order neural network (HONN) can be designed to be invariant to geometric transformations such as scale, translation, and in-plane rotation. Invariances are built directly into the architecture of a HONN and do not need to be learned. Thus, for 2D object recognition, the network needs to be trained on just one view of each object class, not numerous scaled, translated, and rotated views. Because the 2D object recognition task is a component of the 3D object recognition task, built-in 2D invariance also decreases the size of the training set required for 3D object recognition. We present results for 2D object recognition both in simulation and within a robotic vision experiment and for 3D object recognition in simulation. We also compare our method to other approaches and show that HONNs have distinct advantages for position, scale, and rotation-invariant object recognition.The major drawback of HONNs is that the size of the input field is limited due to the memory required for the large number of interconnections in a fully connected network. We present partial connectivity strategies and a coarse-coding technique for overcoming this limitation and increasing the input field to that required by practical object recognition problems.

Effects and correction of magneto-optic spatial light modulator phase errors in an optical correlator

Here we study the optical phase errors introduced into an optical correlator by the input and filter plane magneto-optic spatial light modulators. We measure and characterize the magnitude of these phase errors, evaluate their effects on the correlation results, and present a means of correction by a design modification of the binary phase-only optical-filter function. The efficacy of the phase-correction technique is quantified and is found to restore the correlation characteristics to those obtained in the absence of errors, to a high degree. The phase errors of other correlator system elements are also discussed and treated in a similar fashion.

Path planning using optically computed potential fields

An algorithm for the optical computation of potential field maps suitable for mobile robot navigation is described, and experimentally produced maps and paths are presented. The parallel analog optical computation uses a two-dimensional spatial light modulator on which an image of the potential field map is generated. Optically calculated fields contain no local minima, tend to produce paths centered in gaps between obstacles, and produce paths which give preference to wide gaps. Calculation of 128*128 pixel fields at a few hertz is possible with current technology, and calculation time vs. map size scales favorably in comparison to digital electronic computation.<<ETX>>

Resonance transition radiation X-ray lasers

A free electron laser is proposed using a periodic dielectric and helical magnetic field. Periodic synchronism between the electrons and the optical wave is obtained at the period of the dielectric and not at the period of the helical magnetic field. The synchronism condition and the gain of the new device are derived. The effects on the gain of the new device are derived. The effects on the gain from dephasing and beam expansion due to elastic scattering of the electrons in the periodic medium are included in the gain calculation. Examples of the resonance transition radiation laser and klystron are given. Operation at photon energies between 2.5 and 3.5 keV with net gain up to 12% is feasible using high electron-beam energies of 3 and 5 GeV. Moderate (300-MeV) beam energy allows operation between 80 to 110 eV with up to 57% net gain using a klystron design. In both cases, rapid foil heating may limit operation to a single electron-beam pulse. >

Determining object orientation with a hierarchical database of binary synthetic discriminant function filters

An optical correlation-based system is demonstrated which recognizes an object and determines its angular orientation by traversing a hierarchical database of binary filters. The database architecture is made possible by the development of binary synthetic discriminant function filters.

An empirical comparison of ID3 and HONNs for distortion invariant object recognition

The authors present results of experiments comparing the performance of the ID3 symbolic learning algorithm with a higher-order neural network (HONN) in the distortion invariant object recognition domain. In this domain, the classification algorithm needs to be able to distinguish between two objects regardless of their position in the input field, their in-plane rotation, or their scale. It is shown that HONNs are superior to ID3 with respect to recognition accuracy, whereas, on a sequential machine, ID3 classifies examples faster once trained. A further advantage of HONNs is the small training set required. HONNs can be trained on just one view of each object, whereas ID3 needs an exhaustive training set.<<ETX>>

Electron beam emittance growth in thin foils: A betatron function analysis

The emittance of an electron beam increases due to multiple scattering when passing through one or more thin foils. The effect of a given foil on a beam’s emittance is dependent on whether the beam is diverging, converging, or at a waist. A method for calculating the growth in emittance using betatron functions is presented. The technique provides a full description of the beam in phase space after a thin scatterer.