08.02.2013 Views

New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...

New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...

New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

20 CHAPTER 2. PRELI<strong>MI</strong>NARIES<br />

Figure 2.2.8: Bricks <strong>of</strong> <strong>the</strong> two kinds <strong>of</strong> operas. Some <strong>of</strong> <strong>the</strong>m are labeled with <strong>the</strong><br />

number <strong>of</strong> <strong>the</strong> group <strong>the</strong>y belong to. Notice <strong>the</strong> difference in <strong>the</strong> two bricks <strong>of</strong> group 1<br />

and recall that <strong>the</strong> only condition <strong>for</strong> a group membership is <strong>the</strong> position while o<strong>the</strong>r<br />

properties such as color may be different.<br />

Step 3: Analyze <strong>the</strong> groups<br />

For each group from <strong>the</strong> previous step check if all its bricks roughly look <strong>the</strong><br />

same or if we can identify differences between bricks from an opera-A pile<br />

versus an opera-B pile. For example, within <strong>the</strong> same group bricks from an<br />

opera-A pile might be darker than bricks from an opera-B pile which would<br />

<strong>the</strong>n be two sub-groups with different average color. This is illustrated in<br />

Figure 2.2.9: we are looking at seven random members <strong>of</strong> group 1 (see Figure<br />

2.2.8) and <strong>the</strong>re are obviously two sub-groups: <strong>the</strong> bricks from opera-B type<br />

piles have a white spot so <strong>the</strong> average color <strong>of</strong> a brick in this opera-B subgroup<br />

is around 230. Bricks from an opera-A type pile do not have this white<br />

spot which decreases <strong>the</strong> average brightness value to about 200.<br />

(a) B,<br />

color:<br />

231<br />

(b) A,<br />

color:<br />

205<br />

(c) B,<br />

color:<br />

134<br />

(d) B,<br />

color:<br />

232<br />

(e) A,<br />

color:<br />

204<br />

(f) A,<br />

color:<br />

206<br />

(g) A,<br />

color:<br />

201<br />

Figure 2.2.9: Seven lego bricks <strong>of</strong> group 1 in Figure 2.2.8. Obviously, <strong>the</strong>re exist<br />

two sub-groups.<br />

From now on we will call a group that has two different sub-groups a<br />

feature. So here a feature has three properties: <strong>the</strong> group number and <strong>the</strong><br />

average color values <strong>of</strong> its two sub-groups (<strong>for</strong> group 1: 200 <strong>for</strong> <strong>the</strong> opera-A<br />

type and 230 <strong>for</strong> <strong>the</strong> opera-B type).<br />

Now, <strong>the</strong> usage <strong>of</strong> a feature to determine an unknown pile’s type is quite<br />

simple: From our feature we know <strong>the</strong> average position (since it is still a<br />

group) and <strong>the</strong> color a brick would have if it belongs to an opera-A or opera-B<br />

type pile. Consequently, we just have to take <strong>the</strong> picture <strong>of</strong> <strong>the</strong> unknown pile,<br />

identify <strong>the</strong> brick at this features position and determine its color and compare<br />

it to <strong>the</strong> colors <strong>for</strong> an opera-A or an opera-B type. The closer it is to one <strong>of</strong><br />

<strong>the</strong>m, <strong>the</strong> more likely it is that kind <strong>of</strong> opera.<br />

Step 4: Check <strong>the</strong> feature quality<br />

Use each feature found in <strong>the</strong> previous step to guess <strong>the</strong> label <strong>of</strong> a pile we<br />

already know. If <strong>the</strong> answer is correct, we count this as a success. We now<br />

test each feature with every (already labeled) picture we have and count <strong>the</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!