/[gzz]/manuscripts/Paper/paper.tex
ViewVC logotype

Diff of /manuscripts/Paper/paper.tex

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 1.54 by jvk, Mon Mar 24 17:24:17 2003 UTC revision 1.55 by jvk, Fri Mar 28 12:53:07 2003 UTC
# Line 1242  On a darker background, this approach wo Line 1242  On a darker background, this approach wo
1242  Experiments on black-and-white %(faces,)  Experiments on black-and-white %(faces,)
1243  ink blots and snow crystals  ink blots and snow crystals
1244  \cite{goldstein71visualrecognition} show that  \cite{goldstein71visualrecognition} show that
1245  complex pictures can be remembered and recognized and that recognition  complex, unfamiliar pictures can be remembered and recognized
1246    and that recognition
1247  performance decreases very little over time.  performance decreases very little over time.
1248    
1249  We have conducted a pilot experiment (with one subject)  We have conducted an experiment in a similar setting comparing
1250  in a similar setting comparing  the recognition performance of our textures with solid color backgrounds.
1251  the recognition performance of our textures with solid color backgrounds  Five participants performed the experiment in both conditions
1252  (using colors from the same distribution that is used for the textures).  (in random order).
1253  First, 15 textures were shown sequentially, 5 seconds each.  
1254  Then, recognition was tested by showing the 15 seen textures  15 target backgrounds and 15 distractor backgrounds were randomly
1255  with 15 unseen textures in a random order, and the subject answered  chosen for both conditions.
1256  old or new. The test was repeated with solid color backgrounds  The distribution of the solid colors was the same that is used for
1257  for comparison.  the colors of textures with the difference that the highest
1258    lighness tail was de-emphasized to increase the otherwise low
1259  The results were as follows:  discriminability of very light, unsaturated colors.
1260  for textures, .3 bits per trial of the old/new information got through  
1261  (80 percent correct: 12/15 correctly recognized as seen and  First, the 15 target backgrounds were shown sequentially,
1262  12/15 correctly identified as not seen);  in a random order, 5 seconds each.
1263  for solid colors, only .003 bits per trial got through  Then, recognition was tested by showing the 15 target backgrounds
1264  (53 percent correct: 13/15 correctly recognized as seen and  and 15 distractor backgrounds in a random order and
1265  3/15 correctly identified as not seen). XXX  having the participant answer ``old'' or ``new'' for each one.
1266    The time for answers was not limited.
1267    
1268    The results are summarized below
1269    (percentage of trials and reaction time in seconds):
1270    \begin{center}
1271    \begin{tabular}{c|cc|cc}
1272          & \multicolumn{2}{c|}{textures \% (RT)}  & \multicolumn{2}{c}{colors \% (RT)} \\
1273          & correct  & incorrect & correct & incorrect \\
1274    \hline
1275    old   & 84 (3.0) & 16 (5.6) & 43 (3.4) & 57 (4.0) \\
1276    new   & 72 (2.6) & 28 (5.5) & 71 (2.8) & 29 (4.1) \\
1277    \hline
1278    total & 78 (2.8) & 22 (5.5) & 57 (3.1) & 43 (4.1)
1279    \end{tabular}
1280    \end{center}
1281    
1282  %80, 53 fail to bring out the results correctly. Should  %80, 53 fail to bring out the results correctly. Should
1283  %use mutual information or something like that. In colors, the subject  %use mutual information or something like that. In colors, the subject
1284  %was basically guessing!!!  %was basically guessing!!!
1285    
1286  Thus, the subject was able to recognize textures ($p < .001$)  % RT less for correct responses: [F(1,4) = 5.2, p = .082]
1287  while basically guessing the solid colors.  Thus, the recognition performance of the textures was good while
1288  Of course, the quantitative results have little statistical significance,  the solid colors do not have enough variation for unambiguous
1289  but in conjunction with our experiences, they do suggest that the  recognition (without additional cues).
1290  textures are quite recongizable while solid colors do not have  The observed difference in recognition performance is statistically
1291  enough variation for unambiguous recognition (without additional cues).  significant [F(1,4) = 19.0, p = .012].
1292    
1293  The experiment measured the recognition of only 15 textures,  Of course, the experiment measured the recognition of only 15 textures,
1294  and surely it would be harder to remember, e.g., 100 textures.  and surely it would be harder to remember, e.g., 100 textures.
1295  However, the user does not have to remember all the textures;  However, the user does not have to remember all the textures;
1296  it suffices to learn the textures of the most often used documents.  it suffices to learn the textures of the most often used documents,
1297    see Fig.~\ref{fig-zipf}.
1298  Many studies of web cache statistics  Many studies of web cache statistics
1299  have shown that file popularity approximately follows Zipf's law  have shown that file popularity approximately follows Zipf's law
1300  (see, e.g. [xxxref]) so that a small number of documents accounts  (see, e.g. [xxxref]) so that a small number of documents accounts

Legend:
Removed from v.1.54  
changed lines
  Added in v.1.55

savannah-hackers-public@gnu.org
ViewVC Help
Powered by ViewVC 1.1.26