SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf ·...

17
SE4D03 – Computer User Interfaces The Science of Visualization continued The Science of Visualization continued The final topic in visualization interfaces involves an investigation of the neural processing power involving spatial and orientational tuning and not colour sensitivity, other than luminance. Texture Issues (from C.Ware, Information Visualization: Perception for Design, 2 nd Ed., (Morgan Kaufmann Publishers, 2004) ISBN:1-55860-819- 2.) The model used for textural understanding is the GABOR model, where the receptive field of these neurons can be described by means f Gb f i hi h i d f i i d of a Gabor function which is composed of a cosine wave grating and a gaussian as shown below: + = y x Ox ) ( exp cos C sponse Re 2 2 Where C is an amplitude term (or contrast value), S provides the overall size of the Gabor Function by adjusting both the = S S exp cos C sponse Re overall size of the Gabor Function by adjusting both the wavelength of the cosine grating and the rate of decay of the gaussian envelope, and O is a rotation matrix that orients the cosine wave. se4d03 slide L32-1

Transcript of SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf ·...

Page 1: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

SE4D03 – Computer User Interfaces• The Science of Visualization continued• The Science of Visualization – continued

– The final topic in visualization interfaces involves an investigation of the neural processing power involving spatial and orientational tuning and not colour sensitivity, other than luminance.

– Texture Issues (from C.Ware, Information Visualization: Perception for Design, 2nd Ed., (Morgan Kaufmann Publishers, 2004) ISBN:1-55860-819-g , , ( g , )2.)• The model used for textural understanding is the GABOR model,

where the receptive field of these neurons can be described by means f G b f i hi h i d f i i dof a Gabor function which is composed of a cosine wave grating and a

gaussian as shown below:

⎟⎟⎞

⎜⎜⎛ +−

⎟⎞

⎜⎛=

yxOx )(expcosCsponseRe22

– Where C is an amplitude term (or contrast value), S provides the overall size of the Gabor Function by adjusting both the

⎟⎟⎠

⎜⎜⎝

⎟⎠

⎜⎝

=SS

expcos CsponseRe

overall size of the Gabor Function by adjusting both the wavelength of the cosine grating and the rate of decay of the gaussian envelope, and O is a rotation matrix that orients the cosine wave.

se4d03 slide L32-1

Page 2: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- I• Texture Considerations continued• Texture Considerations – continued

– Properties of the Gabor Function• Texture function works by multiplying the cosine wave grating on the

left below, by the gaussian envelope in the centre to provide the two-dimensional Gabor function shown on the right below

• We say that this example shows an excitatory centre flanked by two inhibitory bars:inhibitory bars:

P h th t f l t f t t i th bilit f th– Perhaps the most useful aspect of texture is the ability of the human visual process to segment or parse visually the size and orientation of all possible textures as shown on the next slide.

se4d03 slide L32-2

Page 3: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- II• Texture Considerations continued• Texture Considerations – continued

– Properties of the Gabor Function• For segmentation, two-dimensional array of Gabor detectors filter g , y

every part of the image for all possible orientations and sizes. Areas exciting particular classes of detectors form the basis of visually distinct segments of the image.

se4d03 slide L32-3

Page 4: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- III• Texture Considerations• Texture Considerations –

continuedProperties of the Gabor– Properties of the Gabor Function• At the left is a series of

columns where the left-hand column illustrates the same cosine pattern paired with p pdifferent gaussian multipliers

• The centre column are textures created using eachtextures created using each Gabor Function by reducing the size by a factor of 5 and spreading it in the fieldspreading it in the field

• The right-hand column shows the 2D Fourier transforms of these textures

se4d03 slide L32-4

transforms of these textures.

Page 5: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- IV• Texture Considerations – continued

– Properties of the Gabor Function• It is interesting to note that the human visual perception filters

seem to obey the Uncertainty Principle which is that given aseem to obey the Uncertainty Principle which is that given a constant density of data, the orientation or size can be precisely specified but not both. (See last slide illustration.) When the gaussian is large the spatial frequency is specified quitegaussian is large the spatial frequency is specified quite precisely (Fourier transform is small), but when the gaussian is small the position is well specified but the spatial frequency is not shown by a large Fourier transformnot shown by a large Fourier transform.

• Note however, that a single Gabor detector of the human visual processing system can still beprocessing system, can still be thought of as being tuned to a little packet of orientation and size information that can be positionedinformation that can be positioned anywhere in space. This provides an excellent opportunity to use texture as a segmentation method for visual

se4d03 slide L32-5

a segmentation method for visual images as shown at the left.

Page 6: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- V• Texture Considerations continued• Texture Considerations – continued

– Properties of the Gabor Function• Segmentation to be optimized means that the receptor• Segmentation to be optimized means that the receptor

channels should carry information that is perceptually independent, or as some say, orthogonal meaning that there is no cross-talk or coupling amongst informationno cross-talk or coupling amongst information.

• For texture, it has been determined that orientation should differ by over 30 degrees and spatial frequencies p qshould differ by at least a factor of 3 or 4 for segmentation purposes g p pusing texture alone.

• This is shown at the left between dotted and

se4d03 slide L32-6

between dotted and solid pairs of lines.

Page 7: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- VI• Texture Considerations continued• Texture Considerations – continued

– Properties of the Gabor Function• Texture Distinguished Example• Texture Distinguished Example

– The word TEXTURE is visible only because of texture differences between the letters and the background, overall l i i h ld t tluminance is held constant.

– At the left:(a) only texture orientation is alteredorientation is altered(b) Texture orientation and size are altered andare altered, and(c) Texture contrast is altered, which is the best manner in whichbest manner in which to distinguish texture changes.

se4d03 slide L32-7

Page 8: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- VII• Texture Considerations continued• Texture Considerations – continued

– Properties of the Gabor Function• Texture Contrast Effects• Texture Contrast Effects

– Textures can appear distorted due to contrast effects– Example: a given texture on a coarsely textured

background will appear finer than the same texture on a finely textured background as shown below, left.

– Here, the two ,patches left of centre and right of centre have the same texture granularity, but texture contrast makes them appear different

se4d03 slide L32-8

Page 9: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- VIII• Texture Considerations – continued

– Properties of the Gabor Function• Texture Orientation Effects

Textures can appear distorted due to orientation effects– Textures can appear distorted due to orientation effects– Example: The two horizontal lines are parallel, but they

appear tilted due to orientation contrast.C l i• Conclusions• We have looked at tradeoffs in

design choices about whether to guse colour , shape, texture, or motion, to display a particular set of variables. They include:y1. The concept of low-level

channels that tell us about coding dimensions where separate channels involve colour, elements of form (orientation and size), position, simple motion and

se4d03 slide L32-9

simple motion and stereoscopic depth.

Page 10: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

The Science of Visualization -- IX• Low-Level Human Vision CharacteristicsLow Level Human Vision Characteristics

– Conclusions – cont’d2. For glyphs (markers) to be seen rapidly, they must stand out clearly

from all other objects in their near vicinity in at least one codingfrom all other objects in their near vicinity in at least one coding dimension

3. Channels can visually interfere with each other; generally “like interferes with like” -- a small set of symbols on a texturedinterferes with like -- a small set of symbols on a textured background, a texture with a grain size similar to that of the symbols will make them difficult to distinguish.

4. To read different data dimensions, each of these values should be4. To read different data dimensions, each of these values should be mapped to a different data dimension. Mapping one variable to colour and another to glyph orientation will make them independently readable. If one variable is mapped to X-direction size and another to Y-direction size, they will be readable as a group rather than independently. Also a set of symbols that are hard to see because of a textured background, they can be made to stand out by using another coding channel or having them oscillate will also make them distinctcoding channel or having them oscillate will also make them distinct.

5. And finally care must be taken to avoid the effects of illusionary properties as discussed in the previous slides.

se4d03 slide L32-10

Page 11: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

GUI du Jour -- I• Text that sucks in CHI (particularly error messages)Text that sucks in CHI (particularly error messages)

• Here “ACM” appears randomly in headings,headings,

• As does “Now“• Not good consistency

se4d03 slide L32-11

Page 12: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

GUI du Jour -- II• Text that sucks in CHI (particularly error messages)

• Again random inconsistencies (on the left)

• Much better approach with labeling text on the right example.

se4d03 slide L32-12

Page 13: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

GUI du Jour -- III• Text that sucks in CHI (particularly error messages)Text that sucks in CHI (particularly error messages)

• Pronouns again not consistent, as notedas noted by circles for a few instances.

se4d03 slide L32-13

Page 14: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

GUI du Jour -- IV• Text that sucks in CHI (particularly error messages)Text that sucks in CHI (particularly error messages)

All ll• All you really need to search for a Jeep dealer is the Zip C dCode

• Trimming the searchsearch space, does not save that much time.

se4d03 slide L32-14

Page 15: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

GUI du Jour -- V• Text that sucks in CHI (particularly error messages)Text that sucks in CHI (particularly error messages)

• This is much better, at the l ftleft.

• But the b tt ibottom is best.

se4d03 slide L32-15

Page 16: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

GUI du Jour -- VI• Text that sucks in CHI (particularly error messages)Text that sucks in CHI (particularly error messages)

• Microsoft on a MAC – really yinformative!.

se4d03 slide L32-16

Page 17: SE4D03 – Computer User Interfaces - McMaster Universityse4d03/lectures.new/4d3lec32.pdf · Design, 2nd Ed.,,( g , ) (Morgan Kaufmann Publishers, 2004) ... • Texture function works

GUI du Jour -- VIIText that sucks in CHI (particularly error messages)• Text that sucks in CHI (particularly error messages)

se4d03 slide L32-17