-
Notifications
You must be signed in to change notification settings - Fork 13
/
cudaMorphology.cu
220 lines (198 loc) · 8.99 KB
/
cudaMorphology.cu
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
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
#include <stdio.h>
#include <iostream>
#include "cudaConvUtilities.h.cu"
#include "cudaMorphology.h.cu"
////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// ***Generic Morphologoical Operation Kernel Function***
//
// This is the basis for *ALL* other morpohological operations. Every
// morphological operation in this library can be traced back to this
// (the optimized 3x3 ops are hardcoded/unrolled versions of this function)
//
// For all morph operations, we use {-1, 0, +1} ~ {OFF, DONTCARE, ON}.
// This mapping allows us to use direct integer multiplication and
// summing of SE and image components. Integer multiplication is
// much faster than using lots of if-statements.
//
// Erosion, dilation, median, and a variety of weird and unique
// morphological operations are created solely by adjusting the
// target sum argument (seTargSum).
//
////////////////////////////////////////////////////////////////////////////////
//
// Target Sum Values:
//
// The following describes under what conditions the SE is considered to "hit"
// a chunk of the image, based on how many indvidual pixels it "hits":
//
//
// Erosion: Hit every non-zero pixel
//
// If we hit every pixel, we get a +1 for every non-zero elt
// Therefore, our target should be [seNonZero]
//
// Dilation: Hit at least one non-zero pixel
//
// If we miss every single pixel: sum == -seNonZero
// If we hit one pixel: sum == -seNonZero+2;
// If we hit two pixels: sum == -seNonZero+4;
// ...
// Therefore, our target should be [-seNonZero+1] or greater
//
//
// Median: More pixels hit than not hit
//
// Since each pixel-hit is a +1, and each pixel-miss is a -1,
// the median is 1 if and only if there are more +1s than -1s.
// Therefore, our target should be [0] or greater
//
//
// THEREFORE, consider the following structuring elt with 6 non-zero pixels:
//
// se = 0 1 0
// 1 1 1
// 1 1 0
//
// Morph_Generic_Kernel(se, targSum = 6) ~ Erode
// Morph_Generic_Kernel(se, targSum = -5) ~ Dilate
// Morph_Generic_Kernel(se, targSum = 0) ~ Median
//
////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
__global__ void Morph_Generic_Kernel(
int* devInPtr,
int* devOutPtr,
int imgRows,
int imgCols,
int* sePtr,
int seRowRad,
int seColRad,
int seTargSum)
{
CREATE_CONVOLUTION_VARIABLES(seRowRad, seColRad);
shmOutput[localIdx] = -1;
const int seStride = seColRad*2+1;
const int sePixels = seStride*(seRowRad*2+1);
int* shmSE = (int*)&shmOutput[ROUNDUP32(localPixels)];
COPY_LIN_ARRAY_TO_SHMEM(sePtr, shmSE, sePixels);
PREPARE_PADDED_RECTANGLE_MORPH(seRowRad, seColRad);
__syncthreads();
int accumInt = 0;
for(int roff=-seRowRad; roff<=seRowRad; roff++)
{
for(int coff=-seColRad; coff<=seColRad; coff++)
{
int seRow = seRowRad + roff;
int seCol = seColRad + coff;
int seIdx = IDX_1D(seRow, seCol, seStride);
int seVal = shmSE[seIdx];
int shmPRRow = padRectRow + roff;
int shmPRCol = padRectCol + coff;
int shmPRIdx = IDX_1D(shmPRRow, shmPRCol, padRectStride);
accumInt += seVal * shmPadRect[shmPRIdx];
}
}
// If every pixel was identical as expected, accumInt==seTargSum
if(accumInt >= seTargSum)
shmOutput[localIdx] = 1;
__syncthreads();
devOutPtr[globalIdx] = (shmOutput[localIdx] + 1) / 2;
}
////////////////////////////////////////////////////////////////////////////////
// Standard 3x3 erosions, dilations and median filtering
// Standard (8-connected) 3x3 morph operations
CREATE_3X3_MORPH_KERNEL( Dilate, -8,
1, 1, 1,
1, 1, 1,
1, 1, 1);
CREATE_3X3_MORPH_KERNEL( Erode, 9,
1, 1, 1,
1, 1, 1,
1, 1, 1);
CREATE_3X3_MORPH_KERNEL( Median, 0,
1, 1, 1,
1, 1, 1,
1, 1, 1);
// 4-connected (cross-shaped) structuring elements
CREATE_3X3_MORPH_KERNEL( Dilate4connect, -4,
0, 1, 0,
1, 1, 1,
0, 1, 0);
CREATE_3X3_MORPH_KERNEL( Erode4connect, 5,
0, 1, 0,
1, 1, 1,
0, 1, 0);
CREATE_3X3_MORPH_KERNEL( Median4connect, 0,
0, 1, 0,
1, 1, 1,
0, 1, 0);
////////////////////////////////////////////////////////////////////////////////
// There are 8 standard structuring elements for THINNING
CREATE_3X3_MORPH_KERNEL( Thin1, 7,
1, 1, 1,
0, 1, 0,
-1, -1, -1);
CREATE_3X3_MORPH_KERNEL( Thin2, 7,
-1, 0, 1,
-1, 1, 1,
-1, 0, 1);
CREATE_3X3_MORPH_KERNEL( Thin3, 7,
-1, -1, -1,
0, 1, 0,
1, 1, 1);
CREATE_3X3_MORPH_KERNEL( Thin4, 7,
1, 0, -1,
1, 1, -1,
1, 0, -1);
CREATE_3X3_MORPH_KERNEL( Thin5, 6,
0, -1, -1,
1, 1, -1,
0, 1, 0);
CREATE_3X3_MORPH_KERNEL( Thin6, 6,
0, 1, 0,
1, 1, -1,
0, -1, -1);
CREATE_3X3_MORPH_KERNEL( Thin7, 6,
0, 1, 0,
-1, 1, 1,
-1, -1, 0);
CREATE_3X3_MORPH_KERNEL( Thin8, 6,
-1, -1, 0,
-1, 1, 1,
0, 1, 0);
////////////////////////////////////////////////////////////////////////////////
// There are 8 standard structuring elements for PRUNING
CREATE_3X3_MORPH_KERNEL( Prune1, 7,
0, 1, 0,
-1, 1, -1,
-1, -1, -1);
CREATE_3X3_MORPH_KERNEL( Prune2, 7,
-1, -1, 0,
-1, 1, 1,
-1, -1, 0);
CREATE_3X3_MORPH_KERNEL( Prune3, 7,
-1, -1, -1,
-1, 1, -1,
0, 1, 0);
CREATE_3X3_MORPH_KERNEL( Prune4, 7,
0, -1, -1,
1, 1, -1,
0, -1, -1);
CREATE_3X3_MORPH_KERNEL( Prune5, 7,
-1, -1, -1,
0, 1, -1,
1, 0, -1);
CREATE_3X3_MORPH_KERNEL( Prune6, 7,
-1, -1, -1,
-1, 1, 0,
-1, 0, 1);
CREATE_3X3_MORPH_KERNEL( Prune7, 7,
-1, 0, 1,
-1, 1, 0,
-1, -1, -1);
CREATE_3X3_MORPH_KERNEL( Prune8, 7,
1, 0, -1,
0, 1, -1,
-1, -1, -1);