Lecture 1 - introduction
human-centered design: combination of discoverability and feedback
trade-off preferred over solution
Lecture 2 - an example of applied research
eye tracking (ET)
assumption: people attend where they look
you don’t have full control over where you look
advantages: good quality
disadvantages: expensive, hard
mouse tracking (MT)
always double task (eye tracking + mouse tracking)
you have full control so MT is decision-making
advantages: cheap, easy
disadvantages: bad quality
experimental design
trade-off between representative enough and good to work with for sample
time targets
→ fixed: predetermined time
→ self-paced: participant chooses to click next target
prompting
providing cues to remember information participant might not recall spontaneously
→ unaided recall: no prompting
→ aided recall: prompting
spatial gaze can be shown in a heat map
temporal aspects of viewing behavior
1. viewing time/ stimulus duration (higher in MT)
2. number of fixations (eye movements; higher for MT)
3. fixation duration (eye movement; longer with MT)
4. dwell time (eye movement; longer with MT)
5. spontaneous and aided recall (higher during MT)
6. message transfer and viewing times (higher during MT)
limitations experiment on computer
1. resolution (pixels and colors)
2. throughput (processing speed and bandwidth)
1
,color photo
three channels RGB
or CMYK photo (Cyan, Magenta, Yellow and blacK)
→ the higher the number of pixels, the more pixels coded with a color, the better the picture
bit depth: each bit can have two values (0 or 1), so with an 8-bit image (RBG) you will have 2^8
= 256 colors per channel
left picture: 8 bits = 256 colors per channel
middle picture: 4 bits = 16 colors per channel
right picture: 3 bits = 8 colors per channel
bit to byte: 8 bits together = 1 byte
1 megabyte (MB) = 1000 kilobytes (kB) = 1.000.000 bytes (B) = 8.000.000 bits (b)
calculate file size: image
pixels length x pixels height x bits (x 3 if colored pic) / 8 = bytes / 1000 = kB
→ 8 bit per channel photo: 8 x 3 (RBG; actually 3 pics in 1 because of 3 colors) = 24 bit
→ colored 8-bit pic 200 x 300 pixels: 200 x 300 x 3 x 8 = 1.440.000 bits / 8 = 180.000 byte /
1000 = 180 kB
reduce black-white photo: code to 1 bit per pixel (black = 0, white = 1)
grayscale: each pixel can have an 8-bit (0-255) or 16-bit (0-65535) code → the larger the
number, the smaller the increments between various shades of gray
human open pose
temporal frequency: images per second in a video
→ 100 Hz = 100 times per second
more pixels per cm = more bits per color = sampling at a higher frequency
result: lots of data to save (hard for computer)
solution: choose temporal sampling frequency that only keeps relevantion information
Nyguist frequency = sampling rate / 2
the Fsample (= device, like CD or camera) must be twice as high as your Nyguist frequency
(=Fsignal, like music)
2
,→ a property of the signal you are trying to digitize (Fsignal), not the device used for sampling
(Fsample)
→ you determine the Fsample (so how quickly it must be sampled to correctly digitize without
losing information) of a signal based on the Nyguist frequency
examples
1. if you use a device to measure with 1000 Hz, the Nyguist frequency (= highest
frequency you can pick up) is 500 Hz
2. the highest frequency a healthy young person can hear is around 22 kHz → sample CD
at 44 kHz (so twice as high) → the Nyguist frequency is 22 kHz (= highest frequency
you can hear from CD)
if Fsample = Fsignal
you don’t capture the signal (you don’t see the bird
moving)
if Fsample is slightly higher than Fsignal
you have a sample problem
moiré artefacts
the outside world always has a higher resolution than any
camera
example: device with 200 Hz can capture frequencies till
100 Hz, otherwise moiré patterns
solution: use low-pass filter before sampling (digitize)
3
, compression: reducing file size by smartly encoding information
more compression = smaller image = more processing power necessary to unpack it
benefits: faster data transmission, reduced storage requirements
drawbacks: loss of quality, compatibility issues, irreversible changes
examples compression methods
image compression: tiff, jpg, bmp, png
sound compression: wav, aiff, mp3
video compression: AVI
lossy compression (JPG, MP3)
reduce file size very effectively by permanently eliminating some of the less important data
→ only use if trade-off between size reduction and loss of quality is acceptable
lossless compression (PNG, TIFF)
stores only the differences between the frames and uses an algorithm to form complete file
advantage: small files without lost information (ideal for experiments)
disadvantage: it takes processing power to decode a compressed file
examples
computer with a lot of computation power but not much memory → use lossless files
computer that is lacking computation power but has a lot of memory → use uncompressed
files
dpi = dots per inch
graphical formats image compression
PNG GIF TIFF JPG BMP
color depth (bits) 48 8 24 24 24
lossy compression yes
lossless compression yes yes yes yes
suitable for experiment ++ - + -- +
4 parameters audio files
1. sampling frequency
higher frequency = large files
for healthy human being the maximum frequency is 22 kHz, so for CD quality you
need 2 x 22 kHz = 44 kHz
4
human-centered design: combination of discoverability and feedback
trade-off preferred over solution
Lecture 2 - an example of applied research
eye tracking (ET)
assumption: people attend where they look
you don’t have full control over where you look
advantages: good quality
disadvantages: expensive, hard
mouse tracking (MT)
always double task (eye tracking + mouse tracking)
you have full control so MT is decision-making
advantages: cheap, easy
disadvantages: bad quality
experimental design
trade-off between representative enough and good to work with for sample
time targets
→ fixed: predetermined time
→ self-paced: participant chooses to click next target
prompting
providing cues to remember information participant might not recall spontaneously
→ unaided recall: no prompting
→ aided recall: prompting
spatial gaze can be shown in a heat map
temporal aspects of viewing behavior
1. viewing time/ stimulus duration (higher in MT)
2. number of fixations (eye movements; higher for MT)
3. fixation duration (eye movement; longer with MT)
4. dwell time (eye movement; longer with MT)
5. spontaneous and aided recall (higher during MT)
6. message transfer and viewing times (higher during MT)
limitations experiment on computer
1. resolution (pixels and colors)
2. throughput (processing speed and bandwidth)
1
,color photo
three channels RGB
or CMYK photo (Cyan, Magenta, Yellow and blacK)
→ the higher the number of pixels, the more pixels coded with a color, the better the picture
bit depth: each bit can have two values (0 or 1), so with an 8-bit image (RBG) you will have 2^8
= 256 colors per channel
left picture: 8 bits = 256 colors per channel
middle picture: 4 bits = 16 colors per channel
right picture: 3 bits = 8 colors per channel
bit to byte: 8 bits together = 1 byte
1 megabyte (MB) = 1000 kilobytes (kB) = 1.000.000 bytes (B) = 8.000.000 bits (b)
calculate file size: image
pixels length x pixels height x bits (x 3 if colored pic) / 8 = bytes / 1000 = kB
→ 8 bit per channel photo: 8 x 3 (RBG; actually 3 pics in 1 because of 3 colors) = 24 bit
→ colored 8-bit pic 200 x 300 pixels: 200 x 300 x 3 x 8 = 1.440.000 bits / 8 = 180.000 byte /
1000 = 180 kB
reduce black-white photo: code to 1 bit per pixel (black = 0, white = 1)
grayscale: each pixel can have an 8-bit (0-255) or 16-bit (0-65535) code → the larger the
number, the smaller the increments between various shades of gray
human open pose
temporal frequency: images per second in a video
→ 100 Hz = 100 times per second
more pixels per cm = more bits per color = sampling at a higher frequency
result: lots of data to save (hard for computer)
solution: choose temporal sampling frequency that only keeps relevantion information
Nyguist frequency = sampling rate / 2
the Fsample (= device, like CD or camera) must be twice as high as your Nyguist frequency
(=Fsignal, like music)
2
,→ a property of the signal you are trying to digitize (Fsignal), not the device used for sampling
(Fsample)
→ you determine the Fsample (so how quickly it must be sampled to correctly digitize without
losing information) of a signal based on the Nyguist frequency
examples
1. if you use a device to measure with 1000 Hz, the Nyguist frequency (= highest
frequency you can pick up) is 500 Hz
2. the highest frequency a healthy young person can hear is around 22 kHz → sample CD
at 44 kHz (so twice as high) → the Nyguist frequency is 22 kHz (= highest frequency
you can hear from CD)
if Fsample = Fsignal
you don’t capture the signal (you don’t see the bird
moving)
if Fsample is slightly higher than Fsignal
you have a sample problem
moiré artefacts
the outside world always has a higher resolution than any
camera
example: device with 200 Hz can capture frequencies till
100 Hz, otherwise moiré patterns
solution: use low-pass filter before sampling (digitize)
3
, compression: reducing file size by smartly encoding information
more compression = smaller image = more processing power necessary to unpack it
benefits: faster data transmission, reduced storage requirements
drawbacks: loss of quality, compatibility issues, irreversible changes
examples compression methods
image compression: tiff, jpg, bmp, png
sound compression: wav, aiff, mp3
video compression: AVI
lossy compression (JPG, MP3)
reduce file size very effectively by permanently eliminating some of the less important data
→ only use if trade-off between size reduction and loss of quality is acceptable
lossless compression (PNG, TIFF)
stores only the differences between the frames and uses an algorithm to form complete file
advantage: small files without lost information (ideal for experiments)
disadvantage: it takes processing power to decode a compressed file
examples
computer with a lot of computation power but not much memory → use lossless files
computer that is lacking computation power but has a lot of memory → use uncompressed
files
dpi = dots per inch
graphical formats image compression
PNG GIF TIFF JPG BMP
color depth (bits) 48 8 24 24 24
lossy compression yes
lossless compression yes yes yes yes
suitable for experiment ++ - + -- +
4 parameters audio files
1. sampling frequency
higher frequency = large files
for healthy human being the maximum frequency is 22 kHz, so for CD quality you
need 2 x 22 kHz = 44 kHz
4
