/[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 1.25 by jvk, Mon Mar 17 13:08:25 2003 UTC revision 1.26 by jvk, Mon Mar 17 16:51:21 2003 UTC
# Line 1077  On a darker background, this approach wo Line 1077  On a darker background, this approach wo
1077    
1078  \subsection{Recognizability and memorizability}  \subsection{Recognizability and memorizability}
1079    
 JVK  
   
 - lots of texture perception work on texture discrimination  
1080    
1081  - we need also memorization; therefore, careful evaluation is needed  JVK
   
 - our textures are on a higher level, more like complete pictures  
1082    
1083  - experiments on black-and-white (faces,) ink blots, and snow crystals  There has been
1084    \cite{goldstein71visualrecognition} show that  lot of texture perception work on texture discrimination.
1085    complex pictures can be remembered and recognized and that recognition  However, in our application texture discrimination is not as
1086    performance decreases very little over time.  much of an issue as memorizability and recognizability of
1087    previously seen textures.
1088  - pilot test: 15 textures or pure colors shown sequentially, 5 seconds each.  Furthermore, our textures are on a higher level,
1089    recognition tested by showing the 15 seen textures mixed with 15 new  more like complete pictures than the usually studied microstructure.
1090    textures in a random order, subject answers old/new.  
1091    Experiments on black-and-white %(faces,)
1092  - results: for textures, 80 percent correct (12/15 correctly recognized  ink blots, and snow crystals
1093    as seen and 12/15 correctly identified as not seen).  \cite{goldstein71visualrecognition} show that
1094    for pure colors, 53 percent correct (13/15 correctly recognized  complex pictures can be remembered and recognized and that recognition
1095    as seen and 3/15 correctly identified as not seen).  performance decreases very little over time.
1096    
1097  - the experiment only measured the ability to dicriminate previosly seen  We have conducted a pilot experiment (with one subject)
1098    textures from unseen textures. However, our experience shows that  in a similar setting comparing
1099    at least the most recurring textures can easily be associated to the  the recognition performance of our textures with pure color backgrounds.
1100    document content.  First, 15 textures (or pure colors) were shown sequentially, 5 seconds each.
1101    Then, recognition was tested by showing the 15 seen textures mixed
1102    with 15 unseen textures in a random order, and the subject answered
1103    old or new.
1104    The results were as follows:
1105    for textures, 80 percent correct (12/15 correctly recognized
1106    as seen and 12/15 correctly identified as not seen);
1107    for pure colors, 53 percent correct (13/15 correctly recognized
1108    as seen and 3/15 correctly identified as not seen).
1109    
1110    Of course, this result is not statistically relevant,
1111    but in conjunction with our experiences, it suggests that the
1112    textures are quite recongizable and that they are clearly
1113    superior to pure colors in terms of recognizability.
1114    
1115    Furthermore, the experiment only measured the ability to
1116    recognize previosly seen textures within unseen textures.
1117    Our experience shows that in practice,
1118    at least the most recurring textures are not only recognized but can
1119    also easily be associated to the document content.
1120    
1121  - as per Zipf's law [XXXref], it suffices to learn the textures  - as per Zipf's law [XXXref], it suffices to learn the textures
1122    of the most recurring documents.    of the most recurring documents.

Legend:
Removed from v.1.25  
changed lines
  Added in v.1.26

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