Perception of Pixelated Images

By Talis Bachmann

Perception of Pixelated Images covers the increasing use of these images in everyday life as communication, socialization, and commerce increasingly rely on technology. The literature in this book is dispersed across a wide group of disciplines, from perception and psychology to neuroscience, computer science, engineering, and consumer science.

The book summarizes the research to date, answering such questions as, What are the spatial and temporal limits of perceptual discrimination of pixelated images?, What are the optimal conditions for maximizing information extracted from pixelated images?, and How does the method of pixelation compromise or assist perception?

• Integrates research from psychology, neuroscience, computer science, and engineering
• Explains how the process of perception works for pixelated images
• Identifies what assists and hinders perception, including the method of pixelation
• Discusses the limits of perception of pixelated images

• Language: English
• Publisher: Academic Press
• Release date: Jan 21, 2016
• ISBN: 9780128095058

Talis Bachmann

Talis Bachmann is a professor in the departments of law and psychology at the University of Tartu in Estonia, specializing in Cognitive and Forensic Psychology. He is also head of the Perception and Consciousness group in the Estonian Center of Behavioral and Health Sciences. He is on the Executive Board of the Union of Estonian Psychologists, and is a member of the Association for Scientific Studies of Consciousness, and Association for Psychological Science. He currently serves on the board of Consciousness and Cognition, was the former co-editor of Acta Universitatis Scientiarum Socialium et Artis Educandi Tallinnensis, and was a former board member on The European Journal of Cognitive Psychology. He is an author of 190 academic publications. Talis is regarded as one of the leading experts in masking, microgenesis, and perception of pixelated visual images.

Book preview

Preface

Talis Bachmann

There are two major world-views, not necessarily mutually exclusive. According to one of these the build of reality is smooth and richness and variations in the world around us and within us as well as the basic processes underlying all reality, are due to the gradients of gradual changes. Diversity is carried by increments and decrements of continuous change. Waves and oscillations are the essence; borders and discontinuities are either ambiguous, represent hidden continuities, or are conventions. According to the other stance, reality with its variations comes down to discrete units, particles, quanta; apparent smoothness and continuity are either an illusion or the result of imperfection of the observational or measurement devices. This dichotomy of attitude or of the practice of approach applies to spatial as well as temporal dimensions of reality. It covers philosophy as well as natural sciences—take the monadology of Leibniz or wave-versus-quantum dispute in physics as only two of the many well-known examples. However, for an experimental psychologist or neuroscientist who studies perception, this juxtaposition may seem a bit too radical. In most cases when research on perception is at stake it is easy to show that both approaches are meaningful. From the objective perspective, the reality of our subject matter allows both, the quantal, discrete-unit-based ways of analysis or approach, as well as the paradigms acknowledging a continuous, gradual organized reality. Receptors and neurons and action potentials can be considered as the quantal reality, easy to describe and analyze as parts of digital systems processing arrays of 1 s and 0 s. Categorical perception or semantic nodes in respective networks can also be conceived of as examples of discreteness. On the other hand, waves of brain activity recorded by EEG or MEG or smooth changes of brightness or color observed in visual psychophysics are empirical examples allowing approaches that acknowledge graduality and smoothness as attributes of subject matter.

In this book we will not dwell on theoretical debates about the continuous or quantum-like nature of the underlying research realities. We just pick up a domain where quantization offers itself robustly at its face value and in a relatively widespread practice of technology and communication—pixelated visual images. Nowadays, pixels are everywhere, be it your computer monitor, digital photography, or a shot of the to-be-witness in criminal cases shown on TV with the facial image pixelated. This is in order to hide the personal identity. In the latter case we have an interesting situation. The original image of a visual object—a social subject—which was initially pixelated with very high resolution is additionally pixelated at a more coarse spatial level of pixelation. The actually pixelated (ie, spatially quantized) original objective image was formed from a spatial array of discrete, quantal changes of luminance, but this was done with such a high spatial density (very high spatial resolution) that subjectively this image is experienced as an area covered or filled with a continuous, smooth change of luminance and contrast. Objective spatial resolution of luminance gradients can be much better than subjective resolution, and the discrete, objective pixel-wise changes become assimilated in subjective perception in a smooth, gradual manner. Now, by applying a special transform aided by a software program on the spatial distribution of the luminous pixels it is possible to average the luminance of these original tiny pixels within larger square-shaped areas and get an image where the—now larger—pixels can be subjectively perceived and the smoothness of contrast gradients is replaced by abrupt, discrete changes of brightness. The ever-discrete objective visual depiction of the original image (carrying image contents) can become both an indiscrete, seemingly continuous change of luminance in space where it comes to subjective perception, or a discrete, quantally changing, abrupt step-like change in luminance also represented subjectively. What matters is whether the spatial scale of pixelation, where neighboring pixels carry different levels of luminance, remains below the subjective threshold of spatial contrast discrimination or not.

As in practice there are plenty of actual or potential circumstances where pixelated images have to be or happen to be perceived, it is of practical and theoretical interest to know what the psychophysical regularities and constraints characterizing the perception of pixelated images are. What the spatial and temporal limits of perceptual discrimination of pixelated images are, what the optimal conditions for maximizing information that can be extracted when perceiving the pixelated images are, how pixelated images can be used in basic research on perception and what kind of scientific knowledge has been obtained by this method; also, how the method of pixelation can be used in increasing the capacity and versatility of communication technology without compromising the minimum necessary quality of perception by human observers—these are some of the typical questions posed in experimental research on the perception of pixelated images.

At present, information about the perception of pixelated images is scattered throughout mutually separate and sometimes even isolated publications belonging to the domains of experimental psychology, neuroscience, engineering, and even art. I decided to write this book mainly because of the wish to offer a source of information where, for the first time, most of the pertinent work is brought together in a single volume. I will review this work and provide some methodological and theoretical comments. The courage to take on this task probably comes from my own experience in this topic area where I have published some experimental studies beginning from the 1980s. Partly the motivation to write this text comes from noticing that pixelation has been often regarded simply as a means of image transform equivalent to low-pass filtering performed in the Fourier or Gaussian domain. This is a mistake that sometimes may be pardoned as it does not have detrimental consequences for interpreting experimental data, but sometimes it does. Thus, I will also focus on differentiating the effects pixelation causes in comparison with other ways of spatial filtering. I wanted to prepare this volume also because pixelated images are a common feature of the habitat of a modern individual living his or her life in the midst of rich and varied stimulation carried by contemporary communication technologies. It should be useful to understand how you perceive what surrounds you.

I see the main audience of this book consisting of graduate and postgraduate students working on the problems of visual perception, (digital) image processing, visual communication, and face perception. It must be easier for them to take one book instead of searching and looking through thousands of scattered sparse sources buried among the mass of available data that is troublesome to access. I mean psychology, neuroscience, and cognitive science students. As the approach and focus of this text is contextuated in psychology and psychophysics of visual perception and cognition, the book may be of interest to IT and communication technologies specialists—both hardware- and software-oriented—who are very well informed about the topics of pixelation from the engineering and computing perspective, but somewhat less knowledgeable about what psychologists have found about this topic. Additionally, a layman who is not afraid of some technical terms and has interest toward experimental psychology may be a welcome reader of this short book.

Several colleagues and students have been an essential support and personally rewarding coworkers in researching on psychological processes by using pixelated stimuli-images. Thank you, Neeme Kahusk, John MacDonald, Søren Andersen, Endel Põder, Iiris Tuvi (Luiga), Laura Leigh-Pemberton, Triin Eamets, Hanne-Loore Härma, Carolina Murd, Kevin Krõm, et al.! I am also really grateful to Nikki Levy, Barbara Makinster, Kiruthika Govindaraju and other Elsevier people who have been very helpful and efficient in making this book a reality. And of course, this is not just a mere traditional cliché when here and now, I express my loving feelings to my family who have always been a permanent support even though I have stolen too much precious time from them in favor of the lab and the pixels staring at me from the monitors of my computers.

Chapter 1

Introduction

Visual Images and How They Are Dealt With

Abstract

To set a broader context for our topic, we use a simple tripartite framework of stages ultimately leading to the process of image perception. First, there are real objects and scenes in the environment (1). As a second stage, an image is produced (taken, captured, formed) and stored representing the characteristics of these objects or scenes (2). Third, a perceiver (human or robot) processes this image and gets some more or less adequate information about the reality represented by its corresponding image (3). To proceed from (1) to (2), some means to capture reality-information must be used. The typical example is a camera—a camera allowing projection of the image on film or a set of electronic sensors capable of feeding a certain system of encoding and storage of visual data. But scanners, radars, etc. can also be used. To proceed from (2) to (3) the captured and stored image has to be presented for the perceiver whose brain performs processing of the signals from this image and builds up cognitive representation of the objects or scenes captured and reflected in the image. Thereby, an idea of the depicted reality is formed in perception. Perception of reality as a subjective (phenomenal) image thus becomes mediated by an image of reality formed by some technical means. In our case of interest, the images are digital images somewhat impoverished by virtue of consisting only of square-shaped local areas (blocks or pixels of low resolution) with uniform intensity within each pixel. And our perceivers are not robots, but human subjects with natural brains capable of visual processing by inborn mechanisms and acquired visual skills.

Keywords

Pixels; images; illuminance; sampling; distortion; compression; image dimensions; quality metrics; visual channels; processing stages; spatial frequency; sine wave grating; Fourier anlysis; point light sources; contrast; modulation transfer function; neuron; receptive field; spatial scale

To set a broader context for our topic, let us use a simple tripartite framework of stages ultimately leading to the process of image perception. First, there are real objects and scenes in the environment (1). As a second stage, an image is produced (taken, captured, formed) and stored representing the characteristics of these objects or scenes (2). Third, a perceiver (human or robot) processes this image and gets some more or less adequate information about the reality represented by its corresponding image (3). To proceed from (1) to (2), some means to capture reality-information must be used. The typical example is a camera—a camera allowing projection of the image on film or a set of electronic sensors capable of feeding a certain system of encoding and storage of visual data. But scanners, radars, etc. can also be used. To proceed from (2) to (3) the captured and stored image has to be presented for the perceiver whose brain performs processing of the signals from this image and builds up cognitive representation of the objects or scenes captured and reflected in the image. Thereby, an idea of the depicted reality is formed in perception. Perception of reality as a subjective (phenomenal) image thus becomes mediated by an image of reality formed by some technical means. In our case of interest, the images are digital images somewhat impoverished by virtue of consisting only of square-shaped local areas (blocks or pixels of low resolution) with uniform intensity within each pixel. And our perceivers are not robots, but human subjects with natural brains capable of visual processing by inborn mechanisms and acquired visual skills.

However, before we can get to the central subject matter of this book—human perception of pixelated images—we need to introduce some basic terms and knowledge about images, technical aspects of image processing and evaluation as well as about how the visual system of the human brain processes visual signals.

1.1 Digital Images and Sampling

The definition of image is well given by Anbarjafari (Video Lectures on Digital Image Processing, University of Tartu): An image is a two-dimensional function f(x,y), where x and y are the spatial (plane) coordinates, and the amplitude of f at any pair of coordinates (x,y) is called the intensity of the image at that level. Here we must only add that in our less abstract context the image is a depiction of reality captured by some technical means. The intensity at a point of space x,y in our case refers to the intensity of visible light. The local intensities of light emitted and/or reflected from real objects or scenes change smoothly (indiscretely) as we move over the space of these objects or scenes. However, "if x,y and the amplitude values of f are finite and discrete quantities, we call the image a digital image. A digital image is composed of a finite number of elements called pixels, each of which has a particular location and value" (Anbarjafari, op. cit.). An image can represent some object, scene, or process from which the information about local intensities is sampled. Sampling can be either a simultaneous process where all local sampled values are measured at once or a successive, step-wise process. Basically, a digital image is a numerical representation of a two-dimensional image. For instance, if you sample information useful for estimating the brightness of a chess board and use zeros (0 s) for black squares and ones (1 s) for white squares, and if you move your sampling steps so that for each square you have two steps of sampling (starting with a black square), luminance values for one row of the chess-board squares will be marked up as 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1. Now, in order to get from a one-dimensional (1D) representation to a two-dimensional (2D) representation you should perform the same procedure on the next row of squares adjacent to the first sampled row, and so on until the whole 2D area of interest is sampled. (The next row then would be described as 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0.) This way we end up with a matrix of intensity values, each having a spatial coordinate x,y. Understandably, the values need not include zero and, equally understandably, we can use more levels of values for characterizing each point on an image instead of the 1-bit example above. (Thus, in the case of an 8-bit sampled representation we have 256 possible values of intensity (eg, luminance) extracted from certain spatial locations.)

It is obvious that digital sampling or digitization means that a digital image is an approximation of the real object or scene it represents. Consequently, we lose information by effecting a sampling procedure. This is unless we could have an ideal sampling device, an ideal sampler, the spatial resolution of which and the spatial progression of whose sampling operation precisely follows the object or scene it attempts to sample for information about it. An ideal sampler would produce samples equivalent to the instantaneous value of the continuous signal at the sampling points of interest. (Fig. 1.1 shows the basic principle of