58fdd08b
 Jiabing Li
added median filter
|
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
|
//#include <thrust/sort.h>
//#include <thrust/execution_policy.h>
//#include <thrust/binary_search.h>
//#include <thrust/device_ptr.h>
template <typename T>
__device__ void cuswap ( T& a, T& b ){
T c(a);
a=b;
b=c;
}
namespace stim{
namespace cuda{
template<typename T>
__global__ void cuda_median2(T* in, T* out, T* kernel, size_t X, size_t Y, size_t K){
int xi = blockIdx.x * blockDim.x + threadIdx.x;
int yi = blockIdx.y * blockDim.y + threadIdx.y;
size_t Xout = X - K + 1;
size_t Yout = Y - K + 1;
int i = yi * Xout + xi;
//return if (i,j) is outside the matrix
if(xi >= Xout || yi >= Yout) return;
//scan the image, copy data from input to 2D window
size_t kxi, kyi;
for(kyi = 0; kyi < K; kyi++){
for(kxi = 0; kxi < K; kxi++){
kernel[i * K * K + kyi * K + kxi] = in[(yi + kyi)* X + xi + kxi];
}
}
//calculate kernel radius 4
int r = (K*K)/2;
//sort the smallest half pixel values inside the window, calculate the middle one
size_t Ki = i * K * K;
//sort the smallest half pixel values inside the window, calculate the middle one
for (int p = 0; p < r+1; p++){
for(int kk = p + 1; kk < K*K; kk++){
if (kernel[Ki + kk] < kernel[Ki + p]){
cuswap<T>(kernel[Ki + kk], kernel[Ki + p]);
}
}
}
//copy the middle pixel value inside the window to output
out[i] = kernel[Ki + r];
}
template<typename T>
void gpu_median2(T* gpu_in, T* gpu_out, T* gpu_kernel, size_t X, size_t Y, size_t K){
//get the maximum number of threads per block for the CUDA device
int threads_total = stim::maxThreadsPerBlock();
//set threads in each block
dim3 threads(sqrt(threads_total), sqrt(threads_total));
//calculate the number of blocks
dim3 blocks(( (X - K + 1) / threads.x) + 1, ((Y - K + 1) / threads.y) + 1);
//call the kernel to perform median filter function
cuda_median2 <<< blocks, threads >>>( gpu_in, gpu_out, gpu_kernel, X, Y, K);
}
template<typename T>
void cpu_median2(T* cpu_in, T* cpu_out, size_t X, size_t Y, size_t K){
#ifndef USING_CUDA
//output image width and height
size_t X_out = X + K -1;
size_t Y_out = Y + K -1;
float* window = (float*)malloc(K * k *sizeof(float));
for(int i = 0; i< Y; i++){
for(int j =0; j< X; j++){
//Pick up window elements
int k = 0;
for (int m=0; m< kernel_width; m++){
for(int n = 0; n < kernel_width; n++){
window[k] = in_image.at<float>(m+i, n+j);
k++;
}
}
//calculate kernel radius 4
r_ker = K * K/2;
//Order elements (only half of them)
for(int p = 0; p<r_ker+1; p++){
//Find position of minimum element
for (int l = p + 1; l < size_kernel; l++){
if(window[l] < window[p]){
float t = window[p];
window[p] = window[l];
window[l] = t; }
}
}
//Get result - the middle element
cpu_out[i * X_out + j] = window[r_ker];
}
}
#else
//calculate input and out image pixels & calculate kernel size
size_t N_in = X * Y; //number of pixels in the input image
size_t N_out = (X - K + 1) * (Y - K + 1); //number of pixels in the output image
//size_t kernel_size = kernel_width*kernel_width; //total number of pixels in the kernel
//allocate memory on the GPU for the array
T* gpu_in; //allocate device memory for the input image
HANDLE_ERROR( cudaMalloc( &gpu_in, N_in * sizeof(T) ) );
T* gpu_kernel; //allocate device memory for the kernel
HANDLE_ERROR( cudaMalloc( &gpu_kernel, K * K * N_out * sizeof(T) ) );
T* gpu_out; //allocate device memory for the output image
HANDLE_ERROR( cudaMalloc( &gpu_out, N_out * sizeof(T) ) );
//copy the array to the GPU
HANDLE_ERROR( cudaMemcpy( gpu_in, cpu_in, N_in * sizeof(T), cudaMemcpyHostToDevice) );
//call the GPU version of this function
gpu_median2<T>(gpu_in, gpu_out, gpu_kernel, X, Y, K);
//copy the array back to the CPU
HANDLE_ERROR( cudaMemcpy( cpu_out, gpu_out, N_out * sizeof(T), cudaMemcpyDeviceToHost) );
//free allocated memory
cudaFree(gpu_in);
cudaFree(gpu_kernel);
cudaFree(gpu_out);
}
}
#endif
}
#endif
|
