Название: The Birth of Computer Vision
Автор: James E. Dobson
Издательство: University of Minnesota Press
Год: 2023
Страниц: 215
Язык: английский
Формат: pdf (true)
Размер: 10.1 MB

A revealing genealogy of image-recognition techniques and technologies.

Today’s most advanced neural networks and sophisticated image-analysis methods come from 1950s and ’60s Cold War culture—and many biases and ways of understanding the world from that era persist along with them. Aerial surveillance and reconnaissance shaped all of the technologies that we now refer to as computer vision, including facial recognition. The Birth of Computer Vision uncovers these histories and finds connections between the algorithms, people, and politics at the core of automating perception today.

Computer vision methods are often paired with Machine Learning and Artificial Intelligence algorithms because they were developed and articulated alongside each other. What we now recognize as computer vision applications informed the design of many popular and important Machine Learning algorithms.

Computer Vision algorithms are frequently invisible; they may seem esoteric or too highly specialized to give them much serious thought, but these technologies are now utterly pervasive. They run in the background, as it were, seeing and increasingly determining the world. These technologies are not just imposing order on the content of images; as prosthetic visual technologies, they are also changing how we see. In the twenty- first century, the distinction between human and computer vision has become increasingly hard to define. This is no accident. These technologies were initially developed in part to help researchers understand human perception. In modeling perception, such technologies provided a new set of metaphors that enabled a redescription of physiological systems. They are more than just models of human vision; they are designed to exceed our visual capacities. Computer vision can make the invisible visible. It can detect the presence of obscured objects. It can augment reality, identifying danger and tracking movement. It can learn from the past to add structure to the present. It can also replicate biases. Most crucially, the birth of computer vision inaugurated new techniques of governance and regimes of visuality that reconfigure our conceptions of perception and observation. These algorithms have become domesticated, in multiple senses of the word. While many of the core computer vision algorithms were initially developed for tracking and predicting the actions of foreign armies, they are now used in digital photo albums, are widely found in social media applications, and are even embedded in automobiles.

The readings of algorithms that follow are drawn from methods inspired by the discourse of computer vision itself. I use these methods to interpret several major computer vision algorithms by iteratively following the steps of the solutions generated by early attempts to automate perceptual tasks. In several of the primary (if not pri mal) scenes and imaginative scenarios that motivated the birth of computer vision, we see a number of machines and discrete computational units altering data and making decisions, but we can also identify a cast of characters, including the operators of these imagined seeing machines, the knowledge workers whose expertise is proposed to be extracted and automated, and the observers of these scenes themselves— that is, the inventors and parents of computer vision.

One major goal of this book is to deconstruct the absolute division that has been drawn between accounts of human and machine vision. The invention of computer vision has produced a displacement of the observing human subject, but this displacement did not remove the human perceptual subject from computer vision.

The computer vision algorithms that I examine in this book provide much of the predictive power behind contemporary face and object recognition applications. These applications have become increasingly common. Versions of these same algorithms come preinstalled on phones and computers. They are used by all levels of government as well as by nongovernmental organizations, including the corporations that manufacture our devices and software. Most contemporary critical work on computer algorithms has focused on what we might call the phenomenological experience of digital devices — that is, what we experience of the digital world when we use computers, smartphones, and other technological artifacts.

Motivated by the ongoing use of these major algorithms and methods, The Birth of Computer Vision chronicles the foundations of Computer Vision and Artificial Intelligence, its major transformations, and the questionable legacy of its origins.

Chapter 1 provides an overview and history of the field of research known as computer vision and the application, developments, and activities arising from the computational manipulation of digital images.

Chapter 2 situates the previously murky origin story of the development of a major class of computer vision techniques within the field of Artificial Intelligence through an examination of Frank Rosenblatt’s Perceptron, an algorithm as well as a physical device.

Chapter 3 analyzes a crucial moment of transformation in the field of computer vision: the movement from what was called automatic photointerpretation to image understanding and finally computer vision proper.

Chapter 4 examines a particular class of low- level computer vision techniques that were brought into being through the Shakey Project, a 1960s- era DARPA- funded research project into computer-driven robotic automatons.

The Coda offers a short, critical reading of the OpenCV computer vision tool kit. This tool kit provides reference implementations of a number of important algorithms, many of which are introduced in the previous chapters, that were either created during the early moments of the field or are the direct descendants of these initial algorithms. There are over 2,500 different algorithms included in the tool kit, although some are subject to restricted uses as a result of existing software patents. There are bindings— that is, abstract interfaces that make these algorithms easier to use and portable to different hardware and software platforms— for several major programming languages, including C++, Java, and Python. OpenCV is at present one of the core enabling technologies for computer vision. It is the foundation on which innumerable applications are built.

Contents: