-
Notifications
You must be signed in to change notification settings - Fork 1
/
SyncWiki-JPEG2000.txt
125 lines (77 loc) · 3.12 KB
/
SyncWiki-JPEG2000.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
Notes from Newspaper Digitised Masters TAB
==========================================
OCR ---
- Use Hough-style algorithm plus typeface?
- Printers used multiple letter widths to aid justification.
- Use couplet analysis to determine language?
- Are candidate letters kept anywhere?
- Seems strange to loose the contiguity of letters - sets of
simple shapes, not blocks.
Layout Analysis
----------
- This 'semantic' segmentation is perhaps harder than the OCR
bit.
QA Of Digitsed Masters
----------
- Use linear Hough over segments to assess warp?
- Take scan at high res, fix problems, and down-sample? e.g.
fixing page roll.
- DPI - 'as small as the smallest stroke' - can this be
estimated?
- Test readability using 'Humans', e.g. Amazon Turk.
- We can look at everything, if we let our users do it, and tell
us if there are problems.
- SSIM and MS-SSIM - we need implementations.
- PSNR - does the value shift with image size, due to the
limitations on the smallest possible amount of change being a
larger percentage for smaller images?
- Noise Reduction:
- Is the noise model scale-invariant? It was a Gaussian,
essentially, with a threshold.
- Does NR+INCOMP beat COMP?
- NR can be tuned locally, i.e. dense regions can be left
untouched.
- Preferable compared with potential complexity of
Multi-resolution files? Fat pixels? ** Acceptable noise: how to
compare image noise with capture noise. How to QA? MP3 problem?
Noise reduction will always remove the data that happens to look
like noise. ok?
Semi-Vector? Multi-bit-depth files?
- Use compression-decompression cycles to analyse stability of
noise?
- JPEG 2000
- Is conversion to Lum/Chrominance/etc lossy? Surely not.
- JPEG 2000 uses scale-invariant (wavelet) transform, but
Newspapers have a characteristic length scale?
- Minimally lossy means accepting floating-point math. Minimal 1
or 2 points of Greyscale changes on 30% of the pixels. Is
float/double/quad well defined?
- As you push up the resolution of a file, how does the LLCOMP
size come out? Stable?
- Is JPEG-LS better? Is it better enough?
- Is Adobe support for JPEG2000 poor? Their code is poor?
- TIFF has weak ICC support?
- JP2 only allows ICC profiles, no ProPhotoRGB, AdobeRGB?
- NO JPEG SUPPORT IN PDF/A-2?!?!? Surely just added
JPEG2000.
- NPDP - Can we have the full list of parameters please?
Capture
----------
- Important to estimate the noise in the capture process, e.g.
ISO noise and (0.2mm) shift was rather high for our camera
setup.
- Take multiple shots and average them? Link to the space
IR-imaging process.
- Can camera or scanner noise be modelled more precisely?
Digitisation Process
----------
- BL: Some confusion over the overall process. e.g. borders
chopped, saving 20% in space, which seems rather high, and may be
w.r.t uncomp TIFF.
- BL: Need the people doing capture and QA to believe in this
stuff, and be able to sort it out. Process is a littel conflicted
at present.
- At a fixed MB size, I'd rather pump up the resolution than keep
the capture noise.
- Push vendors to reduce the noise floor in their capture
hardware?