@@ -35,7 +35,7 @@ which filter should be used for resampling. Possible values are:
35
35
were changed in this version.
36
36
37
37
Bicubic and bilinear downscaling
38
- --------------------------------
38
+ ++++++++++++++++++++++++++++++++
39
39
40
40
From the beginning ``BILINEAR `` and ``BICUBIC `` filters were based on affine
41
41
transformations and used a fixed number of pixels from the source image for
@@ -52,7 +52,7 @@ If you have previously used any tricks to maintain quality when downscaling with
52
52
steps), they are unnecessary now.
53
53
54
54
Antialias renamed to Lanczos
55
- ----------------------------
55
+ ++++++++++++++++++++++++++++
56
56
57
57
A new ``LANCZOS `` constant was added instead of ``ANTIALIAS ``.
58
58
@@ -66,19 +66,19 @@ The ``ANTIALIAS`` constant is left for backward compatibility and is an alias
66
66
for ``LANCZOS ``.
67
67
68
68
Lanczos upscaling quality
69
- -------------------------
69
+ +++++++++++++++++++++++++
70
70
71
71
The image upscaling quality with ``LANCZOS `` filter was almost the same as
72
72
``BILINEAR `` due to a bug. This has been fixed.
73
73
74
74
Bicubic upscaling quality
75
- -------------------------
75
+ +++++++++++++++++++++++++
76
76
77
77
The ``BICUBIC `` filter for affine transformations produced sharp, slightly
78
78
pixelated image for upscaling. Bicubic for convolutions is more soft.
79
79
80
80
Resize performance
81
- ------------------
81
+ ++++++++++++++++++
82
82
83
83
In most cases, convolution is more a expensive algorithm for downscaling
84
84
because it takes into account all the pixels of source image. Therefore
@@ -95,7 +95,7 @@ The upscaling performance of the ``LANCZOS`` filter has remained the same. For
95
95
times.
96
96
97
97
Default filter for thumbnails
98
- -----------------------------
98
+ +++++++++++++++++++++++++++++
99
99
100
100
In Pillow 2.5 the default filter for :py:meth: `~PIL.Image.Image.thumbnail ` was
101
101
changed from ``NEAREST `` to ``ANTIALIAS ``. Antialias was chosen because all the
@@ -105,7 +105,6 @@ other filters gave poor quality for reduction. Starting from Pillow 2.7.0,
105
105
uses supersampling internally, not convolutions.
106
106
107
107
Image transposition
108
- ^^^^^^^^^^^^^^^^^^^
109
108
110
109
A new method ``TRANSPOSE `` has been added for the
111
110
:py:meth: `~PIL.Image.Image.transpose ` operation in addition to
@@ -127,7 +126,7 @@ implementations use Gaussian blur internally, all changes from this chapter
127
126
are also applicable to it.
128
127
129
128
Blur radius
130
- -----------
129
+ +++++++++++
131
130
132
131
There was an error in the previous version of Pillow, where blur radius (the
133
132
standard deviation of Gaussian) actually meant blur diameter. For example, to
@@ -138,7 +137,7 @@ If you used a Gaussian blur with some radius value, you need to divide this
138
137
value by two.
139
138
140
139
Blur performance
141
- ----------------
140
+ ++++++++++++++++
142
141
143
142
Box filter computation time is constant relative to the radius and depends
144
143
on source image size only. Because the new Gaussian blur implementation
@@ -150,7 +149,7 @@ second for radius 1, 3.6 seconds for radius 10 and 17 seconds for 50, now blur
150
149
with any radius on same image is executed for 0.2 seconds.
151
150
152
151
Blur quality
153
- ------------
152
+ ++++++++++++
154
153
155
154
The previous implementation takes into account only source pixels within
156
155
2 * standard deviation radius for every destination pixel. This was not enough,
0 commit comments