[Gzz-commits] manuscripts/xupdf article.rst

[Janne V. Kujala]

CVSROOT:        /cvsroot/gzz
Module name:    manuscripts
Changes by:     Janne V. Kujala <address@hidden>        03/02/04 07:56:07

Modified files:
        xupdf          : article.rst 

Log message:
        paper hw impl

CVSWeb URLs:
http://savannah.gnu.org/cgi-bin/viewcvs/gzz/manuscripts/xupdf/article.rst.diff?tr1=1.31&tr2=1.32&r1=text&r2=text

Patches:
Index: manuscripts/xupdf/article.rst
diff -u manuscripts/xupdf/article.rst:1.31 manuscripts/xupdf/article.rst:1.32
--- manuscripts/xupdf/article.rst:1.31  Tue Feb  4 01:42:29 2003
+++ manuscripts/xupdf/article.rst       Tue Feb  4 07:56:06 2003
@@ -144,7 +144,7 @@
   they show thumbnails, or what they show).
   Emphasis seems to be on the methods for generating the f+c structure,
   i.e., finding landmarks, paths between nodes, what nodes to show, etc.
-  Some features use in the example views:
+  Some features used in the example views:
 
        - nodes are shown as cubes (focus node as a sphere) with titles
        - node size/height/color depicts importance/access frequency
@@ -302,7 +302,7 @@
 
 When a link is followed, the torn shape of the target buoy 
 animates to the full shape of the document, as if larger and 
-larger parts were torn off.
+larger parts were torn off. [XXX: screenshots]
 
 The hardware implementation uses a noise texture for creating 
 the variation in the torn shape with texture coordinates tied 
@@ -345,38 +345,53 @@
 Paper
 -----
 
-- fragments of documents are shown
+Although the silhuettes of the buoys are different,
+the fragments of the documents still seem quite similar.
+The user could identify the related documents by
+reading the text of a fragment, but that requires attention.
 
-    - fragments of PDF files seem similar; target fragment
-      not useful per se
-
-    - using a unique background texture for each, situation
-      dramatically changes.
-
-A unique background texture consisting of random shapes 
-and colors is chosen for each document using the identity as a seed. 
-The background textures help the user perceive the identity
+Using a unique background texture for each document changes the 
+situation dramatically: the user can perceive the identity
 of the most familiar documents at a glance, 
 even when only fragments are shown.
+Furthermore, when moving from node to node, the pre-attentive
+cues of identity help the user maintain a sense of direction.
+
+The background textures are randomly chosen using the identity as a seed. 
+That is, each document has a unique backround texture, but the texture 
+is not in any way related to the contents of the document (except
+that a hash of the contents could be used as an identity of an immutable
+document).
+That way, the textures in any view are as different as possible,
+even if the documents are similar.
+Furthermore, the unique background of any document can be instantly drawn,
+as soon as the identity inside the system is known.
 
 The distribution of the textures is designed to be maximally 
-diverse and recognizable using a rough, qualitative model 
+diverse and recognizable with respect to a rough, qualitative model 
 of visual perception.
-For example, backgrounds with random pixels would all
+For example, backgrounds with random pixels (noise) would all
 look the same, because the pixels are not perceived individually.
-However, shapes and overall colors can be independently random to 
+However, shapes and overall colors should be independently random to 
 maximize diversity.
 Making the backgrounds repeating creates well-defined patterns
 and improves recognizability when fragments are shown.
 
-- readability
-- preattentive
+Our hardware implementation (libpaper) uses a small set
+of *basis textures*, which are non-linearly combined on the 
+GPU to create a large set of recognizable shapes. 
+The coordinates of the component textures are 
+randomly chosen affine functions of the paper location,
+but repeating with a randomly chosen *repeating unit* 
+(a parallelogram).
+
+At each pixel, the combined values of the basis textures 
+are used for interpolating between the colors
+of a small, randomly chosen palette of *compatible* colors. 
+That way, the colors and shapes are independently random 
+and the palette can be restricted to light colors to 
+maintain readability. 
 
-- hw accelerated impl (libpaper)
-       - colors
-       - basis textures
-       - texcoords, repeating units
-       - combiners
 
 Implementation on the Gzz platform
 ==================================