Full project report - The Department of Computer Science

advertisement
Epistemological Aspects of Computer Vision
via Objectivism
A Project in Introduction to Computational and Biological Vision
By: Gai Boscoboinik
Ori Grosz
031516446
036401263
Introduction
In his book "A guided tour of computer vision", Vishvjit S. Nalwa writes:
"It has now been well over two decades since several individuals and groups first made
concreted efforts to automate visual perception, and yet a sense of frustration seems to prevail.
… Perhaps the impatience is justified. Perhaps, computer vision research has indeed been
painstakingly unproductive; or, perhaps, the working framework adopted by the research
community is hopelessly flawed. "
It has now been well over another decade since Nalwa wrote these words, and yet, no
meaningful advancements have been made in the field of computer vision. At least, none that
are even remotely close to the desirable human perceptual capabilities. Could it be that the
working framework adopted by the research community is indeed flawed? Nalwa does not
expand on this, he only provides in his book an account of the current state (at least, the 1993
state) of computer vision.
The main reason considered by scientists for the relative failure in computer vision research is
that what is known about the human visual system beyond the eye is largely speculative. The
study of the brain is not advanced enough to allow good coherent understanding of the human
visual system. Trying to emulate a system that we can't understand by making assumptions and
abstracting is not very prominent to succeed.
Nalwa considers another reason, but disregards it rather quickly, probably since it is more of a
philosophical issue, rather than a scientific one:
"Before we proceed any further, however, let us pause for a moment to consider whether
perception has any meaning in the absence of what is commonly understood to be intelligence;
such a consideration is especially pertinent given that computer vision has its origins in a field
called artificial intelligence.
Is visual perception an integral component of what we commonly term intelligence? Perhaps
not. Although many researchers subscribe to the view that perception is inextricably linked to
cognition – that what we perceive is often a consequence of logical reasoning…
… Although the question of what constitutes intelligence is of considerable intellectual import,
it has no direct bearing on our discussion here; hence, we shall not delve into it further."
He did not delve into it further. But we will.
The absence of cognition in machines just might be the flaw overlooked by the research
community while trying to make computers "see".
The purpose of this project is to consider and debate on the importance of cognition in visual
perception, to study through philosophical thinking the possibility of a machine to be able to
perceive without it, and to try and discover other potentially philosophical mistakes in the
basics of computer vision that might be the reason for its slow frustrating development.
Definitions
As this is a partially philosophical discussion and is probably being read by computer scientists
whose knowledge of philosophy can be summed up to knowing how to make fun of
philosophers, we shall now describe two important definitions:
Epistemology - Epistemology is the branch of philosophy that studies knowledge. It attempts
to answer the basic question: what distinguishes true (adequate) knowledge from false
(inadequate) knowledge? Practically, this questions translates into issues of scientific
methodology: how can one develop theories or models that are better than competing theories?
It also forms one of the pillars of the new sciences of cognition, which developed from the
information processing approach to psychology, and from artificial intelligence, as an attempt
to develop computer programs that mimic a human's capacity to use knowledge in an
intelligent way.
Objectivism - Objectivism is the name of a unique system of philosophy formulated by Ayn
Rand and presented in her famous novels Atlas Shrugged (nominated second most influential
book in history by the Library of Congress, after the Bible) and The Fountainhead.
Objectivism is an integrated, hierarchical system of philosophy, covering its 5 “classical”
branches: metaphysics, epistemology, ethics, politics and esthetics. Its fundamental
metaphysical and epistemological ideas are the ones relevant to our work:
1. Reality exists as an objective absolute—facts are facts, independent of man's
feelings, wishes, hopes or fears.
2. Reason (the faculty which identifies and integrates the material provided by man's
senses) is man's only means of perceiving reality, his only source of knowledge, his
only guide to action, and his basic means of survival.
Objectivism emphatically rejects materialism, recognizing consciousness as a fundamental,
natural, non-mystical, this-worldly attribute of man.
After reading these definitions and the introduction, the reader can now fully understand what
we mean by "Epistemological aspects of computer vision via objectivism" and can choose
whether this interests him enough to read on.
Approach & Method
Considering the philosophical study that needed to be done in order to reach groundbreaking or
at least some kind of conclusions, we were faced with a huge amount of theories and different
philosophical approaches to the subject at hand.
The issues of perception and\or consciousness alone are enough to write an infinitely amount
of books about in any given philosophy.
Being amateur Objectivists as we are, we naturally chose to tackle the issues through
Objectivism rather than trying to read other theories that are basically wrong to begin with (as
they are based on "voodoo" philosophies).
And so we began with the basics – readings. We needed to make sure our claims are accurate,
based on facts and not speculation of what we think we know. This goes both to our knowledge
of Objectivism and of our knowledge in computer vision. To achieve this, we relied heavily on
both Ayn Rand's and Robert Efron's writings on perception, and on Ohad Ben Shahar's lectures
in the ICBV course.
In addition, throughout the semester, during the "Introduction to computational and biological
vision" lectures we took notes of possible epistemological mistakes, and discussed them. Some
of those were indeed after consideration tossed aside as they were found to be not
epistemological mistakes but rather a misconception by us of the material. Others, such as
"Illusions – an erroneous perception of reality" were obvious errors that needed to be dealt
with.
We then used our acquired knowledge, our logic, reasoning, our long night brainstorming
meetings, to write an academic paper entitled "On the need for an Epistemological Foundation
for Biological and Computer Vision" which is the result and summation of our hard work.
This paper mainly struggles to provide evidence for the importance of further study in the
epistemological aspects of computer vision, and gives elaborate explanations in this subject,
plus a section regarding the erroneous conception of visual illusions, which is a very good
example of an epistemological error.
It is our hope, that this paper will be the foundation of a new type of research into computer
vision, and possibly revolutionize the field. However, as many scientists consider philosophy
to be a joke and feel that it has no part in real research, we are not keeping our hopes up.
Results and Conclusions
The paper we wrote "On the need for an Epistemological Foundation For Biological And
Computer Vision" consists of the following parts:
1. Introduction – In which we describe what computer vision is, its problem, and our
approach to it.
Two theoretical sections:
2. The Philosophy of Materialism – In which we present one of the main
philosophical/epistemological problems we believe to be detrimental to the advance of
computer vision.
3. Consciousness – In which we properly define what consciousness is and discuss its
levels, the perceptual and conceptual.
Three "practical" sections in which we utilize the material from the two theoretical
sections:
4. The principle of reduction and its effects on science – In which we give a concrete
example of erroneous scientific method resulting from an improper epistemology.
5. Vision and Intelligence – In which we discuss the connection between vision and what
is known as intelligence to try and determine if they are inextricably linked.
6. The Epistemological Status of Visual Illusions – In which we demonstrate a common
epistemological mistake in computer vision through the subject of illusions.
The main Conclusions resulting from the paper and our research are these:
1. Vision and intelligence are two entirely different phenomena.
Perception is not a result of logical inference - it is its base. All processes of thought
and all of man’s knowledge are conceptual, but are based on perceived data obtained
via the senses. “Intelligence,” in essence, denotes a man’s ability to deal with
abstractions, and “abstractions” means concepts. The correct hierarchy is: sensations,
percepts, and concepts—in this order.
2. Achieving the ultimate goals of "computer vision" based on a biological model of
visual perception would require extensive knowledge of the nature of consciousness knowledge which does not exist at present.
For more extensive information and elaborate conclusions please read our paper.
Future work
Those who wish to use this work and\or expand it might note that there is a lot of work still to
be done. We have actually opened here a new almost unexplored realm of thoughts, trying to
link philosophy with computer vision.
Re-evaluation of basic principles used in the field of computer vision is extremely
recommended, trying to create new tools based on different epistemological approaches, as is
the further debate on what is perception and how can we better understand our conceptual
process.
Bibliography
Ben Shahar, Ohad - Lectures on "Introduction to Computational and Biological Vision",
BGU computer science department, autumn semester 2006.
Efron, Robert (1967), “Biology Without Consciousness—And Its Consequences,” in Ayn
Rand (Editor), The Objectivist, Vol. 7, Feb. 1968. Reprinted from Perspectives in
Biology and Medicine, Vol. II, No. 1, Autumn 1967, pp. 9-36.
Efron, Robert (1968), “What Is Perception,” in Boston Studies in the Philosophy of
Science, Vol. IV, D. Reidel Publishing Co.
Nalwa, Vishvjit S. "A guided tour of computer vision", Addison-Wesley Pub (February
1993)
Peikoff, Leonard (1993), Objectivism: The Philosophy Of Ayn Rand, Meridian.
Rand, Ayn (1990), Introduction to Objectivist Epistemology, Expanded 2nd edition,
Meridian, New York.
Rand, Ayn (1964), “The Objectivist Ethics,” in The Virtue Of Selfishness, New American
Library, New York.
Rand, Ayn (1996), Atlas Shrugged, New American Library, New York.
Download