Leon D. Harmon

Masking in Visual Recognition: Effects of Two-Dimensional Filtered Noise

It is difficult to recognize portraits that have been coarsely sampled and quantized. Blurring such images improves recognition. A simple, straightforward explanation is that high-frequency noise introduced by the sampling and quantizing must be removed by low-pass filtering to improve the signal-to-noise ratio and hence signal detectability or recognition. Experiments reported here, suggested on the basis of a different model, show instead that noise bands that are spectrally adjacent to the pictures spectrum are considerably more effective in suppressing recognition.

Neuromimes: Action of a Reciprocally Inhibitory Pair

Two electronic neuron models (neuromimes) connected to a common excitatory input and having both self- and mutual-inhibition were used to simulate the pulse patterns which control wing musculature in certain insects. With variation of stimulus frequency, firing patterns changed in discrete steps. Pattern change was found to be hysteretic: the pattern elicited by a given stimulus frequency depended on whether that frequency was approached from above or below. Pattern selection could be controlled by the injection or deletion of a single pulse in the stimulus pulse-train as well as by smooth frequency change.

Studies with artificial neurons. I. Properties and functions of an artificial neuron.

Summary1.A brief outline of the evolution of neuron modeling is given and an argument is presented for studying neural behavior by “black-box” logical equivalence.2.An electronic circuit incorporating many of the digital and analog properties of neurons is described. Having such properties as variable threshold, summation, all-or-none output, absolute and relative refractoriness, and inhibition, it exhibits a considerable amount of functional equivalence to biological structures.3.Properties of the model are described in detail and relevance to biological neuron measurements are shown. Such behavior as time-intensity trade, repetitive firing and temporal summation are readily achieved and are shown to approximate in vivo behavior.4.The model is sufficiently flexible so that with simple parameter changes and external circuit configurations, a wide variety of neurological phenomena can be exhibited and studied closely. By adding stimulus-derived inhibition for example, accommodation and adaptation are obtained.5.Experiments with the model have suggested new relationships which nervous structures may exhibit. A linear dependence of burst pulse number on accommodation time constant, and a summation-division phenomenon are examples of such findings.6.Models of this sort have utility not only for studying single unit properties but also for investigating group interactions. Such studies may be relevant to elucidation of neural network behavior.

Automatic recognition of print and script

A deceptively simple kind of optical pattern recognition deals with print and script. What seemed at one time to be a fairly easy problem area in automated reading of line-like patterns has turned out to be difficult and expensive. The evolution and present state of the art of machine recognition of print and script is examined. On-hand systems relieve large amounts of human drudgery, and both theory and engineering design have advanced greatly in response to pressures caused by our paper explosion. But major problems of cost and effectiveness still exist.

Auditory Nerve: Electrical Stimulation in Man

Auditory perceptions produced in a person deaf to acoustic stimulation were studied by electrically exciting the auditory nerve through permanently implanted electrodes. Pulsed current as small as 1 microampere peak-to-peak could be perceived. Pitch, as reported by the subject, varied with electrode selection, current amplitude, and pulse repetition rate from about 70 to at least 300 pulses per second. Loudness increased with amplitude and duration of pulse stimuli, and to a lesser extent with repetition rate. The total range in amplitude of the stimulus, from threshold to an uncomfortable loudness, was 15 to 20 decibels. Simultaneous stimulation in separate electrodes produced a number of complex effects.

Studies with artificial neurons. III. Mechanisms of flicker-fusion.

Summary1.A simple model is described which establishes a relationship between physiological and psychophysical flicker-fusion data. The three components of the model are a transducer, a filter and a flicker-detector. The model is consistent with the view that fusion is purely retinal.2.An electronic analog is used to realize the model. Its essential elements are a source of generator pot1ential, a low-pass filter, and a simulated ganglion cell.3.The model was originally designed to reproduce the psychophysical results obtained by de Lange. Special importance is attached to the envelope of his curves; the model reproduces this envelope. His low-intensity results are duplicated accurately, while the high-intensity, high-frequency data are derived somewhat less accurately. The high-intensity low-frequency data are attributed to a mechanism not included in the model.4.In the simulation of physiological data obtained by Enroth, close resemblance to the firing characteristics of retinal ganglia in cat is shown by the model. Firing occurs for bursts lasting for at most half the period of the alternating stimulus, the number of spikes per burst diminishing to unity as “fusion” is approached.5.The phase lags and latencies of responses in the model satisfy neurophysiological evidence. Both off-latency and inhibition-latency are satisfactorily reproduced.6.The model has been used to simulate two-component flicker data. The “flicker-detector” employed leads to results very similar to those obtained psychophysically. The phase characteristics of the assumed filter agree less well.

Computer-produced grey scales

Pictorial output from computers is usually binary-that is, microscopically black orwhite at a point owing to the physical processes involved. A number of techniques have been developed in the past to derive subjective grey values using binary-output devices. Several new extensions of these techniques are reported here. The basic procedure is to quantize space into cells which are filled with different sizes and shapes of black areas on a white background (or vice versa). We include several examples of novel and artistic effects which result from judicious selection of patterns or from rules which divide cells into particular classes of black and white regions. Some of the results raise intriguing questions about the physiology and psychology of vision.

Studies with Artificial Neurons, IV: Binaural Temporal Resolution of Clicks

A psychophysical experiment performed by Guttman, van Bergeijk, and David in 1960 showed that binaural auditory resolution of repetitively presented, closely spaced clicks improves as repetition rate is increased. We propose a model in which the action of a single neuron can account for the phenomenon; it depends on a self‐inhibition function that serves to vary temporal resolution with stimulus rate. Single‐spike (click) stimuli elicit output bursts of variable duration; burst lengths are controlled by an output‐derived feedback whose level depends on stimulus repetition rate. An electronic model of a neuron, simulating a cochlear‐nucleus unit, accurately replicates the essential features of the psychophysical data. Two time constants, estimated by extrapolation, are postulated for single units in the human auditory system.

Neural analogs

Information processing in the nervous system is receiving increasing attention from researchers in the communications sciences. Stimulating and effective liaison between neurophysiologists and engineers is apparent on several fronts and is expanding rapidly.

Backward Enhancement of Binaural Fusion

When a proximal pair of clicks in one ear is “probed” by a single contralateral click, with all clicks of the same amplitude, the probe produces two types of binaural fusion: (1) direct fusions with the individual clicks and (2) fusion with an aftereffect of the first click occurring roughly 1 msec after the direct first‐click fusion. The second type is remarkable because it exists even though the second click occurring perhaps more than a millisecond after the aftereffect‐fusion time cannot produce fusion. The interval of masking of the second click decreases with increasing triad‐repetition rate. A neurophysiological model was recently proposed to account for this, but it did not account for the aftereffect fusion. In an attempt to test an improvement of the model, an experiment was run to secure more data on the aftereffect fusion, particularly with respect to its dependence on repetition rate. The results indicate that the aftereffect fusion rarely attaches to an unpaired click, regardless of repetiti...