Introduction

Having started out as a right is what it is like to do is to I got my god what is it about a bit of a and studied the second section of the book “principles of cognitive psychology” under the heading of “perception and recognition”, I am beginning to wonder how much of cognitive psychology overlaps with the study of artificial intelligence. In this blog, I am going to give a summary or an overview of what I have read and understood so far, and then compare it with my personal experience in working with electronics and artificial intelligence. I will therefore divide this blog up into two parts: A) perception and recognition in psychology, and B) image processing and identification in electronics and artificial intelligence. As with previous blogs, this article will conclude with a short reflective paragraph together with plans on future reading.

Part one: perception and recognition in psychology

To begin with, the book describes how the pathways of light starts from the object entering into the eyes and finally reaching the cortex in the brain. This part of the discussion is no different to what we learn at secondary school in physics and biology. The book then moves on to discuss the various visually related phenomen such as: simultaneous contrast, dark adaptation, colour processing, motion processing, visual illusions, pattern recognition and object recognition. For each of these phenomena, there is a summary of proposed the theories along with their supporting evidence. However, as noted in previous blog, these theories are not definitive as we shall see in this blog.

Simultaneous contrast

Simultaneous contrast refers to situations where a certain colour looks darker or lighter depending on the colours of the background. It was suggested that our eyes and brain are more sensitive to the relative difference of illumination than the actual colour itself. This theory has been used to explain how a person can be colour blind and yet able to perceive visually.

Dark adaptation

Dark adaptation is where visibility increases in the dark over a period of time. Unsurprisingly, the enlargement of the pupils was discussed in the book. Although a wide range of experiments proving the existence of dark adaptation are described in the book, there are no further explanation offered. At this stage, I began to question whether I am reading physics and biology or psychology. Nonetheless, just for the sake of amusing myself and out of the determination to find something new, I continued reading.

Colour processing

Colour processing as its name implies is concerned with our ability to distinguish different colours. The book is in the dangling out of the brain scans will use together to study this topic. It has been noted that certain area of the brain known as V4 had an increase of 13% of blood flow when a coloured stimulator is presented. This increase of blood flow in the area is considered as evidence of brain activity in order to process information triggered by the stimulator.

Motion processing

Motion processing is to do with identifying moving objects. I was surprised to find that different parts of our brain are responsible for seeing stationary objects and moving objects. There is an area of the brain commonly known by the profession as V5, which is responsible for identifying all moving objects. Again, brain scans and monitoring of blood flow were used to identify the area. Interestingly, patients with brain damage in area V5 cannot see moving objects. In other words, stationary objects can be seen normally, just like everyone else. However, as soon as the objects begin to move they become invisible. Some patients have reported difficulty in pouring tea or coffee into a cup, because the fluid appeared to be frozen. They cannot gauge when to stop pouring, because they simply cannot see the fluid. In order to gauge accurately, they have to guess when to stop pouring, stop, wait, and look inside the cup when there is no movement. Then, and only then, can they decide accurately whether they need to stop pouring tea or coffee.

Visual illusions

The book then moves on to discuss how visual illusions work. Again, many theories were proposed, but none are definitive. For instance, the Ponzo illusion was explained by three-dimensional vision. In this theory, it is suggested that our brain automatically interprets two-dimensional images as a representative of the three-dimensional objects. In the case of this illusion, the lines that represent a train track gives the illusion that object A is further away from us then object B.  One therefore assume that if both objects are of the same size, the one further away from us would be smaller when presented in two-dimensional format. Since this picture shows that both objects are about the same size on a two-dimensional plane, one would assume that object  A (the one further away) is actually bigger. Of course, this will not be the case if one was to measure it on paper, which is two-dimensional.

However, if this theory was true, one would expect our judgement in three-dimensional situations to be perfect. However the book spine experiment shows that visual illusions occurs in three-dimensional situations as well. In this experiment, three books are placed so that the spines of the books are equally spaced between them all. The two books on the left are opened up facing each other while the third book is opened up facing away from the other two books. Although performed in three-dimensional situation, we arrived at the illusion that the two book facing each other much closer than the other book. Once again, this book finishes this section of discussion without definitive theory or explanation to the phenomenon.

Pattern recognition

Pattern recognition is concerned with recognising single patterns such as letters and  numerical symbols. When there are so many variations in orientation, typeface, size, and writing styles, how does the brain actually recognise these patterns? One theory suggests that the brain uses different for each pattern when a given pattern matches a template, a recognition is said to have occurred. In support of this theory, computational experiments had been set up to demonstrate the use of templates for pattern recognition. For instance, it has been shown that computers can correctly recognised 69% of numerical digits with only a handful of different templates installed per digit. With increasing number of templates per digit, higher accuracy of recognition is also achieved.

However, this method of recognition does not account for variations in writing styles. Therefore, an alternative theory known as feature theory was proposed. According to this theory, we recognise patterns by identifying key features of the pattern. For example, the letter ‘A’ can be described as two straight legs and a connecting crossbar. However it has been shown that word patterns of letters were made up of smaller letters, recognition of the letters is not always consistent with the theory. Once again, the book exits the discussion without a concrete theory in place.

Object recognition

Object recognition is concerned with how we identify objects. According to the theory of recognition by components, we recognise objects by studying the edges of the object. For instance, by looking at the curvature, combination of parallel lines, symmetry and straight lines, we associate what we see with the objects we know. In support of this theory, simple pictures were drawn and subjects were asked to identify these objects. Gradually though, various features of these pictures were removed and success rate in identifying these objects were recorded. It was found that information pertaining to the edges of the objects were crucial to successful identification, whereas those pertaining to surface features such as decoration and patterns make no difference to the performance.

Part two: image processing and identification in electronics and artificial intelligence

While I was working as an electronics engineer, I was working on a project dealing with CCTV footage. The goal of the project was to use the camera footage to monitor a warehouse. Although it would seem to be a simple task, we were asked to design a program that could control up to 50 cameras to home in into problem spots from different angles. It is hoped that by doing so, the company would stand a better chance of capturing crucial features of the intruder from different angles. However, this is easier said than done. For example, one could argue that all we need is a motion sensor. If the movement is detected, then move nearby CCTV to focus on the detection spot. However, what if a cat passes by? Or what if there are more than one person involved in breaking into the premises? How will the system cope? Therefore, our task involved studying CCTV images and identifying a way of recognising common objects.

In this case, the CCTV images are equivalent to retinal images. These images on its own have no meaning whatsoever, or at least not until the brain or the computer has processed the information and interpreted it. In this respect I find what I have been reading in cognitive psychology remarkably similar to what may be called artificial intelligence.

Part of the work that I did, we use edges for object identification. More specifically, we have looked at significant changes in colours to identify edges. The underlying assumption here is that the edges of an object can be identified by colours. For example, a white car that is parked in the middle of the park would have green surroundings around it. Therefore, if we draw a line between colour contrasts, we would be able to trace the shape of the car.

Identifying the shape of a car or any other object was not too difficult. Since many of these objects have common dimensions and size, engineers can use 3-D vectors and are the mathematical tools to identify objects. However, before these mathematical tools can be employed, first we need to know the dimensions that of the objects. Of course, this means interpreting two-dimensional images. This again is a problem. Take a book for example, if the book was placed near the camera, it will look bigger; if however the book is placed further away from the camera, it will look smaller. This is where the study of aspect ratio comes in. Typically, engineers will include objects with known size in their images so that they can calculate the size of the object of interest by comparison. In our project, we made sure that all our cameras can see certain fixed size objects that will help us in our calculation.

In terms of motion detection, we have tried to compare images captured five seconds apart. For the most part, where things are stationary they should not be any differences between the two images. However, if a person is walking by, we should be able to detect the motion by comparing the images before and after. By comparing and generalising the changes observed, our program was able to detect motions and predict future movements.

Reflection

As I began reading this book, I was very worried because I did not know where I was going. It appeared to me that I was not learning anything new or other than basic secondary school level of physics and biology. However, as I read on, I am pleased that my patience paid off. On reflection, I have found the explanation of motion perception and object recognition remarkably similar to the work I have done in the field of electronics and artificial intelligence. I am pleasantly surprised with what I found.

At this point, I am faced with the dilemma. I am absolutely fascinated by the similarity of cognitive psychology and artificial intelligence. Of course, with my background in electronics, I am inclined to investigate this area further. However, it seems to me that doing so will be defeating the objective of this exercise. For this reason, I have decided to stick with this book and just make my own observation and draw my own parallels with electronics and artificial intelligence as I go along.

Moving on

Looking at the table of contents of the book, the next few sections will be dealing with memory, languages and decision-making. No doubt there will be a lot of similarity with what I’ve done before, but until I finish reading the book and summarising my findings on this blog, I shall refrain from going back into my comfort zone.

Perhaps worth noting at this point, in the back of my mind, I am beginning to wonder how much of this overlap with what I personally know as electronics and artificial intelligence is also an overlap with web science.