Vision visionary

Francis Crick's characteristically pithy comment on the back cover is entirely apt: "a unique book by a unique person". Bela Julesz has been director of the laboratory for vision research at Rutgers University, which he joined in 1989 after working for 30 years at Bell Laboratories and where the vast majority of the scientific work referred to in the book was carried out. Julesz was born in Hungary and studied as an electrical engineer working on microwave radars and the encoding of television signals. He left for the United States during the uprising of 1956 and was immediately offered a post at Bell, working on information encoding. Soon he began to read the psychological literature and to make increasingly major contributions to it. By 1964 he was appointed to head the newly created department of sensory and perceptual processes.

Julesz summarises his research in dialogue 11: (i) human stereo vision is a global process, that is image driven with no mobilisation of knowledge (it is a "bottom-up" process); (ii) effortless texture discrimination is a local process based on local conspicuous features that he dubs "textons"; and (iii) "if we add focal attention to the bottom-up processes of early vision, then the many other enigmatic top-down processes, from semantic memory to the symbolic manipulations of cognition, are not necessary in the great majority of perceptual tasks." His work is best known for the introduction of two novel experimental paradigms: random textures and random dot stereograms, both of which adorn the book. A stereo viewer is provided to help the reader appreciate the latter.

The book consists of two sets of seven "dialogues": the first set on general topics, the second set on specific topics concerning the author's research. The first set of dialogues is accessible to anybody with an interest in science and mind. They discuss the enterprise of vision research, the creative process, the role of theories in what the author calls psychobiology, the relationship between mathematics and human psychology, the "mind-body problem without metaphysics", metascientific problems, and plasticity of the mind. The second set of dialogues is primarily addressed to vision researchers.

Each dialogue is between the author (A) and his alter ego (B). As real dialogues go, these are very one-sided, with B relegated to a few comments and A always having the last word. Sometimes, B is a critical alter ego, especially after passages that are particularly self-serving; but equally often B acts as a respectful assistant, prompting A for some aspect of his research that he "appears" to be overlooking. Sometimes, B simply congratulates A. But there is a lot of irritating repetition. One is told the same story a half dozen times (eg Young's theory of trichromacy was anticipated by Palmer, or that Julesz realised the importance of stereo for camouflage breaking).

In the first dialogue, entitled "The enterprise of vision research", Julesz makes two points. First, "perhaps the greatest tribute to a scientific discovery is when people suddenly regard it as a fact of their own culture, so that it becomes a 'scientific folk song', regardless of who invented or discovered it." It is suggested that random dot stereograms are already such a folk song, though the quotation is curiously at odds with the story later in the book that the author complained that the first edition of the influential PDP book did not give proper credit to him for such stereograms. Second, the author lays great store by the master-student relationship, noting that he has worked with a stream of carefully chosen postdoctoral research- ers. After describing the scientific relationship he enjoyed with them, he refers to two of them as "dishonest", because they did not publish papers on their joint research with the author, a curious word indeed when we are told one page later that we should not take ourselves too seriously doing science!

The second dialogue, on the creative process, examines two processes dubbed "conjugacy" and "scientific bilingualism", and the author introspects on how they may have played a part in his research. Conjugacy means reformulating a problem in different terms so that it is tractable; but we are not given any hint of how such terms may be chosen or why nominally the same problem becomes suddenly easier. Scientific bilingualism amounts to the observation that if you are familiar with two fields of endeavour, sometimes the methods of one can be applied to the problems of the other.

The second set of dialogues opens with "some strategic questions about visual perception". The reader is reminded of David Hilbert's remarkable address to the International Congress of Mathematicians in Vienna in 1900 which posed a number of outstanding problems spanning an extraordinary range of issues in pure mathematics and which loomed large over the field for the entire 20th century. Julesz poses 39 problems in early vision, that are highly variable in level of detail, and each of which is based on some fundamental contribution that the author has made to their elucidation. There is no mention of motion, form, colour, or the control loops that play a key role in seeing. Nevertheless, the problems posed are fascinating and their solution would mean a major stride forward in psychobiology. Otherwise, perhaps the most interesting dialogue for students of vision is the one entitled "Foundations of Cyclopean perception revisited", a re-evaluation of the ground-breaking book that the author published in 1971.

The book is part valedictory, part autobiographical, and part putting on record the author's claim to have originated or anticipated many of the developments of early vision over the past 35 years, from cooperative algorithms for stereo, to texture, motion, and even expert systems! It is a remarkable mixture of insight into early vision, technical summary of the current state of play in stereopsis and texture discrimination, thoughts about the processes of doing science, personal reminiscences and homages to friends and colleagues, and all blended with a large amount of personal trumpet blowing.

Julesz states repeatedly that monographs are superior to research papers, noting that Helmholtz's ground-breaking Handbook(s) of Physiological Optics gave insight not only into Helmholtz's researches into vision but into Helmholtz's ways of thinking. It is hard to avoid the thought that Julesz set out to write a book of the same ilk. History alone will tell whether or not Julesz will be judged as being of the same stature as Helmholtz. For now, the most that can be said is that this is a unique book by a unique individual.

Michael Brady is BP professor of information engineering, University of Oxford.