1 --- !ditz.rubyforge.org,2008-03-06/issue
2 title: Consider merging approx and frame types (into 'frame'?), with formats {bayer} x {weighted}, and resolution for libale-{17,19}.
3 desc: Formats would be (RGB, RGBG, BGRG, GBGR, GRGB) X (WEIGHTED, UNWEIGHTED). (defaults?)
7 reporter: David Hilvert <dhilvert@auricle.dyndns.org>
10 creation_time: 2009-02-11 21:48:13.489277 Z
13 id: e9fed8b2d39321d520c90e584bc26f132ae355d8
15 - - 2009-02-11 21:48:14.138242 Z
16 - David Hilvert <dhilvert@auricle.dyndns.org>
19 - - 2009-02-11 21:53:55.346332 Z
20 - David Hilvert <dhilvert@auricle.dyndns.org>
22 - Naming such a merged data type 'image' would probably better match ALE terminology.
23 - - 2009-02-12 10:17:57.149386 Z
24 - David Hilvert <dhilvert@auricle.dyndns.org>
27 Perhaps better than combining the types would be incorporating new formats
28 (esp. bayer) into the approximation type; in particular, the approximation type
29 is always linear, and hence naturally linearly accumulative, whereas the frame
30 type is generally not, making combination of the two perhaps somewhat less than
33 One approach to combination of the types, that may obviate these concerns,
34 would involve disallowing any accumulative operations for certain
35 configurations (e.g., non-floating point types, non-zero black level, and
37 - - 2009-02-12 13:15:42.775264 Z
38 - David Hilvert <dhilvert@auricle.dyndns.org>
41 One advantage to having a combined type (aside from the obvious Irani-Peleg case)
42 would be naturally allowing loads into approximation types from client code (as this
43 behavior would be required in any case for frame types).
45 Prominent challenges to combination of types would probably include ensuring adequate
46 predicate checks and providing an adequate range of functions for handling each of
48 - - 2009-02-12 14:44:42.686169 Z
49 - David Hilvert <dhilvert@auricle.dyndns.org>
52 Note that certain interface elements of frames (esp. bayer patterns) are analogous
53 to patterns (esp. sub-pixel positioning and rendering) that might be eventually
54 useful for approximation types, so that a combination might be useful for this reason
55 also, in addition to other reasons earlier noted.
56 - - 2009-02-12 18:28:38.872208 Z
57 - David Hilvert <dhilvert@auricle.dyndns.org>
60 One approach to resolving the predicate check problem might be to make images
61 have strictly linear properties (black=0, linear gamma), and to introduce
62 non-linearities as part of a transformation object operating on an image. In
63 this way, there would be no restrictions on operations such as accumulation,
64 so that operations allowed for frames and approximations would become roughly
66 - - 2009-02-12 18:34:03.722666 Z
67 - David Hilvert <dhilvert@auricle.dyndns.org>
70 In addition to black and gamma, exposure might also be moved from frame into
72 - - 2009-02-12 22:21:57.753142 Z
73 - David Hilvert <dhilvert@auricle.dyndns.org>
76 Note that bayer patterns (and any sub-pixel positioning) might be moved into
77 transformation, and that the frame type might have a simpler set of varieties
78 (perhaps between 1-channel and 3-channel types).
79 - - 2009-02-13 12:15:27.474904 Z
80 - David Hilvert <dhilvert@auricle.dyndns.org>
83 On the topic of moving structures into transformation, PSF might also be
84 associated with transformation (perhaps with adjustable parameters for
85 determining frame properties, including possibly focus and shake -- although
86 the former should probably eventually be incorporated into 3D code).
87 - - 2009-02-16 18:11:12.360337 Z
88 - David Hilvert <dhilvert@auricle.dyndns.org>
89 - closed with disposition fixed
91 Types under consideration have been merged, with certain parameters migrated
92 as suggested. Further incorporation of parameters (e.g., PSF) would probably