Structure of human eye and vision

Loading...

来自 National Research University Higher School of Economics 的课程

Deep Learning in Computer Vision

42 个评分

National Research University Higher School of Economics

Deep Learning in Computer Vision

42 个评分

课程 5（共 7 门，Specialization Advanced Machine Learning）

Deep learning added a huge boost to the already rapidly developing field of computer vision. With deep learning, a lot of new applications of computer vision techniques have been introduced and are now becoming parts of our everyday lives. These include face recognition and indexing, photo stylization or machine vision in self-driving cars. The goal of this course is to introduce students to computer vision, starting from basics and then turning to more modern deep learning models. We will cover both image and video recognition, including image classification and annotation, object recognition and image search, various object detection techniques, motion estimation, object tracking in video, human action recognition, and finally image stylization, editing and new image generation. In course project, students will learn how to build face recognition and manipulation system to understand the internal mechanics of this technology, probably the most renown and oftenly demonstrated in movies and TV-shows example of computer vision and AI.

从本节课中

Introduction to image processing and computer vision

Welcome to the "Deep Learning for Computer Vision“ course! In the first introductory week, you'll learn about the purpose of computer vision, digital images, and operations that can be applied to them, like brightness and contrast correction, convolution and linear filtering. These simple image processing methods solve as building blocks for all the deep learning employed in the field of computer vision. Let’s get started!

Short introduction to computer vision4:40

Digital images3:35

Structure of human eye and vision6:15

Color models15:57

与讲师见面

Anton Konushin
Senior Lecturer
HSE Faculty of Computer Science
Alexey Artemov
Senior Lecturer
HSE Faculty of Computer Science

Next, I will talk a little bit about human vision,

to describe the difference between the human eyes and human vision systems,

and a digital camera.

Human eye works similar to the pinhole camera model.

We have our own lenses,

which can change shape by using ciliary muscles

to focus on objects at different distances.

We have iris, which is colored annulus with radial muscles.

And we have our own sensory in form of a number of photoreceptor cells,

rods and cones in the retina.

Photoreceptor cells are connected to optical nerve with help of other neuron cells.

But, what we really see in the world around us is very

different with what is captured by digital camera.

The reason is that, our spatial resolution is very non-uniform,

we have higher resolution,

but in the very small area, one angular minute.

To imagine what one angular minute is,

you can look at the thumbnail on the extended hand.

Only this small thumbnail is viewed in higher resolution.

The rest of the view field is in lower resolution.

Our eyes focuses on specific points in the scene,

this process is called fixation for approximately 100 milliseconds,

then our eyes rapidly moves to another point in this thing.

This movement is called the saccade.

So, our eyes capture a set of sparse images in different parts of our surroundings.

The reason why our spatial resolution is very nonuniform,

lies in nonuniform distribution of photoreceptor cells in the retina.

There are two types of photoreceptor cells.

First are called cones,

and the second is called rodes.

Rods are sensitive to the total intensity of light.

Cones can measure color.

And most of the cones are located in a very small area which is called fovea,

in the center of the visual field.

The rods are distributed uniformly in the retina,

but a number of rods are connected to a single neuron.

So, a number of a photoreceptor cells outside the fovea produces a single number.

And in fovea, each photoreceptor cell is connected to a specific neuron,

and pass its own signal.

The non-uniform form of distribution of cells leads to very interesting effect.

During the night, if you look at the sky,

you can see the sea of stars.

There are a lot of stars around you and you feel all of them.

It's very different from when we are looking to the sky during the day.

The reason for this is that, during night,

the cones do not work properly because they require a larger intensity of light.

And the rods are sensitive to light,

even in low intensity range.

So when we're looking at the sky at night,

our cones are not working, and the rods are working.

And because the rods are distributed evenly throughout the retina,

then all our retina is working,

and we see all the stars.

It's a very interesting sensation.

And the next time you look at the sky,

you can try to imagine how your eyes are working,

and understand why you obtain this feeling.

The movement of eyes,

the succession of fixation points depends on the mental task.

For example, if you try to estimate

the material circumstances of the family in this picture,

we look through all things in the room.

Our eyes are fixated on all objects.

If we try to answer other question,

for example to estimate the age of people,

then our eyes will fixate for a long time to each of the many faces.

And if we want to estimate how long the visitor was away from the family,

then our eyes will constantly change the fixation point.

We will fixate on one man's face,

then rapidly move to the other man's face to try to mentally compare the emotions.

So, by only looking at how man's eyes are looking in the picture,

we can predict what his mental task currently is,

to understand what he's thinking about.

Try to imagine what our eyes are really seeing.

We can create a small video by combining images from fixation points.

Recenter each frame at the fixation point.

The result will look like a very shaky,

like a video from a very shaky camera,

much worse than you see in action movies for example.

But, when we look at the world,

we see that it is stable and at constant resolution.

This is actually an illusion which is created by

our mind from various parts fixation points in the surrounding world.

A lot of mind tricks

and psychological illusions are based on the properties of human eyes.

You can find a lot of funny videos in the Internet,

and test these illusions by yourself.

The eyes are only the first part of a human vision system.

From eyes, information is propagated

to the back of the brain for the low level processing,

where edge detection is performed.

Then information is moving ahead and

it passes over the areas of high level object recognition.

There are specific parts in our brain that is devoted,

for example to face recognition,

to place recognition, and so on.

Separate narrow stream from low level vision to

high level object description is devoted to motion information processing.

After high level recognition,

the signal moves to the decision making area in frontal part of the brain.

And approximately after 150 milliseconds from eyes,

the information can reach the motor control center,

in they react on the input visual stimuli.

探索我们的目录

免费加入并获得个性化推荐、更新和优惠。

Coursera 致力于普及全世界最好的教育，它与全球一流大学和机构合作提供在线课程。

© 2018 Coursera Inc. 保留所有权利。

通过 App Store 下载

通过 Google Play 获取