24. Proceedings of the National Electronics Conference, Vol. V, Chicago, September 26, 27, 28, 1949, Haywood Publishing Company, Lafayette, Indiana.

25. Saito, S. and Kato, K., "Statistical Properties of the Fundamental Frequencies on Japanese Speech Voices," Journal of Acoustics of Japan, pp. 111-116, 1958. (In Japanese)

26. Sakai, T. and Inoue, S., "New Instruments and Methods for Speech Analysis, " Journal of the Acoustical Society of America, Vol. 32, No. 4, pp. 441-450, April 1960.

27. Schouten, J. R., "The Perception of Pitch," Philips Technical Review, Vol. 5, No. 10, 1940.

28. Tamiya, J. and Hiramatsu, K., "Some Characteristics of Sliced Speech and Their Applications," Journal of Acoustics of Japan, pp. 143-150, 1958. (In Japanese)

29. Wang, W. S. Y. and Peterson, G. E., "Segment Inventory for Speech Synthesis," Journal of the Acoustical Society of America, Vol. 30, No. 8, pp. 739-742, 743-746, August 1958.

30. Wiren, J. and Stubbs, S. L., "Electronic Binary Selection System for Phoneme Classification," Journal of the Acoustical Society of America, Vol. 28, No. 6, pp. 1082-1091, November 1956.

WADD TR 60-600

TECHNICAL SESSION III

PHYSICAL ANALOGS OF
BIOLOGICAL COMPONENTS

AND SUBSYSTEMS

Moderator: Mr. Louis A. de Rosa
International Telephone and Telegraph

IMAGE PROCESSING AND FUNCTIONAL RETINA SYNTHESIS

E. E. Loebner

RCA Laboratories, Princeton, N. J.

Introduction

This paper is an attempt to employ a single point of view to three, thus far little related, research activities: 1. the operational methodology of image processing, 2. the inquiry into the mechanics of vision and 3. the design of electronic image processing equipment. This single point of view is based on a "systems" approach which, at least during analysis, does not discriminate between biological and synthetic systems. Our interest in the three topics was promoted by the impending fundamental change in the nature of the man-machine relationship, which is being brought about by the accelerated trend toward automation.

If we scrutinize the role of machines, we find that man has attempted to use them as buffers interposed between himself and his environment. Starting from relatively simple rigs, which were operated either manually or in a manner which would release their predetermined action, he has gradually increased their complexity. Nevertheless he has retained the prime control over the machine actions. Man's communication with the machine is two-fold, he sets its controls and reads its indicators. (These functions can be considered independent from the normal input and output functions of the machines.)

Except for a small class of present day automata, the two-way communication between man and machine, has some peculiarly asymmetrical and non-reciprocal aspects. Thus for instance the number of indicators of various sorts exceeds significantly in most cases the diversity and number of activating controls. Man uses a smaller number of access points, that carry messages simultaneously from him to the machine, while a much larger number of message pathways is used in the oposite direction. An example of such asymmetry is the pilot-aircraft relationship: the pilot constantly monitors many more instruments on the panel, than he simultaneously operates knobs, dials, push-bottons, sticks and pedals.

It appears that this currently prevailing asymmetry derives from the man-limitations and man-oriented design approach. With increasing equipment complexity the balance between man and machine performed tasks has been shifting toward the latter. It is man which has become a buffer between the machine and its input environment. The continuation of the current trend in automation requires a two-fold change in the nature of the machine-man links: firstly a replacement of switches, push-bottons, dials etc. by equipment capable of sensing environmental changes and of extracting control signals from stimuli; and secondly the substitution of increased machine realibility and adaptivity for the multitude of indicators. In this paper we have concentrated on the former requirement.

The machine, decoupled from the operational man-machine link, can now be constructed along machine rather then human oriented engineering principles. A significant increase in the number of simultaneous (parallel) inputs becomes possible. This points to a novel design of input equipment, which is no longer operationally man-limited. A very large number of simultaneous input variables can be sensed in complex sensors. These will be capable of processing the impinging stimuli by extracting that portion of the signals, which has been selected by the designer on the basis of optimum criteriality for the desired performance of the machine.

In a quite general sense, such processing, which transforms a multitude, i.e. a spatial distribution, of some physical properties, into a related spatial distribution of the same or different physical properties, we shall term "image processing". Formally, this transformation can be identified with the mathematical concept of mapping. A mapping function can be described either by a specified mapping procedure or by a physical structure which, interacting with a "physical image", will carry out such mapping transformation. If we limit ourselves to optical and electronic images* we can analyse biological visual systems as well as electronic equipment which process data in pictorial form.

The paper has been divided into three sections. The first section deals with current image processing methods. We shall limit ourselves to a predominantly descriptive discussion of the various schemes employed. They emphasize pattern recognition and lean heavily on methods derived from programmed digital computers that have been more or less adapted to handle pictorial input data. There has been general agreement that the analysis of images into classifiable invariant features has so far been limited to the intuitive insight of the investigators and machine designers. No selfcontained theory on the human or machine discriminability of the various image features exists at this time. A search for basic data needed for such a theory leads one to the topic dealt with in the second section of this paper, since knowing the detailed structure and function of neuroretinas could lead to a discovery of the vision language, i.e. the machine language of the visual apparatus, that is part of most biological systems.

Recently, reliable evidence has been accumulating in the biological discipline that vertebrate retinas carry out image processing functions of categorization and sorting based on spatial, temporal and hybrid attributes

The use of electronic refers to any of the many electron properties such as electric, magnetic, conductive etc.