图像的傅里叶变换
#include"bmp.h" #include<cmath> #include<cstring> #include<cstdio> #define PI 3.1415926 //说明:对输入图像进行快速傅立叶变换,要求输入图像的宽和高必须是2的幂次方 void Bitmap::fourier() { int lineByte = (width_p*bitCount / 8 + 3) / 4 * 4; //申请输出图像缓冲区,并初始化为0 unsigned char* m_pImgDataOut = new unsigned char[lineByte*height_p]; memset(m_pImgDataOut, 0, width_p*height_p); //申请傅立叶缓冲区,并初始化为0 ComplexNumber* m_pFFTBuf = new ComplexNumber[width_p*height_p]; memset(m_pFFTBuf, 0, sizeof(ComplexNumber)*width_p*height_p); //输入图像数据进行二维傅立叶变换 ImgFFT2D(dataBuf, m_pImgDataOut, m_pFFTBuf); delete[] m_pFFTBuf; delete dataBuf; dataBuf = m_pImgDataOut; } //说明:对复数结构体数组arrayBuf进行一维快速傅立叶变换,变换后的结果仍存回arrayBuf中 void Bitmap::FFT1D(ComplexNumber *arrayBuf, int n) { //循环变量 int i, k, r; //申请临时复数缓冲区buf1,长度为n ComplexNumber *buf1 = new ComplexNumber[n]; //将arrayBuf拷贝进buf1 memcpy(buf1, arrayBuf, sizeof(ComplexNumber)*n); //申请临时复数缓冲区buf2,长度为n ComplexNumber *buf2 = new ComplexNumber[n]; //将arrayBuf数组元素基2抽取并重新排列 //若0、1、2、3、4、5、6、7八点序列对调后变作0、4、2、6、1、5、3、7 int t1, t2; for (r = 1; pow(2, r)<n; r++){ t1 = pow(2, r); t2 = pow(2, r - 1); for (k = 0; k<t1; k++){ for (i = 0; i<n / t1; i++){ buf2[k*n / t1 + i].real = buf1[k / 2 * n / t2 + i * 2 + k % 2].real; buf2[k*n / t1 + i].imag = buf1[k / 2 * n / t2 + i * 2 + k % 2].imag; } } memcpy(buf1, buf2, sizeof(ComplexNumber)*n); } //采用蝶型算法进行快速傅立叶变换 //buf1是第r级的输入,buf2存放第r级的输出 float c, s; for (r = 1; pow(2, r) <= n; r++){ t1 = pow(2, r); for (k = 0; k<n / t1; k++){ for (i = t1 / 2; i<t1; i++){ //加权因子 c = cos(-2 * PI*(i - t1 / 2) / t1); s = sin(-2 * PI*(i - t1 / 2) / t1); buf1[k*t1 + i].real = buf2[k*t1 + i].real*c - buf2[k*t1 + i].imag*s; buf1[k*t1 + i].imag = buf2[k*t1 + i].imag*c + buf2[k*t1 + i].real*s; } } for (k = 0; k<n / t1; k++){ for (i = 0; i<t1 / 2; i++){ buf2[k*t1 + i].real = buf1[k*t1 + i].real + buf1[k*t1 + i + t1 / 2].real; buf2[k*t1 + i].imag = buf1[k*t1 + i].imag + buf1[k*t1 + i + t1 / 2].imag; } for (i = t1 / 2; i<t1; i++){ buf2[k*t1 + i].real = buf1[k*t1 + i - t1 / 2].real - buf1[k*t1 + i].real; buf2[k*t1 + i].imag = buf1[k*t1 + i - t1 / 2].imag - buf1[k*t1 + i].imag; } } //第r级的输出存入buf1,作为下一级的输入数据 memcpy(buf1, buf2, sizeof(ComplexNumber)*n); } //傅立叶变换的结果存入arrayBuf memcpy(arrayBuf, buf2, sizeof(ComplexNumber)*n); //释放缓冲区 delete[]buf2; delete[]buf1; } //说明:图像数据二维快速傅立叶变换将图像数据变成复数形式,进行两个一维快速 // 傅立叶变换,变换后的频谱以图像形式存入imgBufOut,此处要求图像宽和高 // 都为2的幂次方 void Bitmap::ImgFFT2D(unsigned char* imgBuf,unsigned char *imgBufOut,ComplexNumber* m_pFFTBuf) { //循环变量 int i, j, u, v; //图像数据变成复数类型存入m_pFFTBuf for (i = 0; i<width_p*height_p; i++){ m_pFFTBuf[i].real = imgBuf[i]; m_pFFTBuf[i].imag = 0; } //申请ComplexNumber结构体数组,长度为height_p ComplexNumber *array = new ComplexNumber[height_p]; //先纵向一维快速傅立叶变换 for (u = 0; u<width_p; u++){ for (v = 0; v<height_p; v++){ array[v].real = m_pFFTBuf[v*width_p + u].real; array[v].imag = m_pFFTBuf[v*width_p + u].imag; } FFT1D(array, height_p); for (v = 0; v<height_p; v++){ m_pFFTBuf[v*width_p + u].real = array[v].real; m_pFFTBuf[v*width_p + u].imag = array[v].imag; } } delete[]array; //再横向一维快速傅立叶变换 for (v = 0; v<height_p; v++){ FFT1D(m_pFFTBuf + v*width_p, width_p); } //将频谱图以图像形式存入imgBufOut float t; int i0, j0; for (i = 0; i<height_p; i++){ //i0 = i; //j0 = j; for (j = 0; j<width_p; j++){ if (i<height_p / 2) i0 = i + height_p / 2; else i0 = i - height_p / 2; if (j<width_p / 2) j0 = j + width_p / 2; else j0 = j - width_p / 2; t = sqrt(m_pFFTBuf[i0*width_p + j0].real*m_pFFTBuf[i0*width_p + j0].real + m_pFFTBuf[i0*width_p + j0].imag*m_pFFTBuf[i0*width_p + j0].imag); t = t / 500; if (t>255) imgBufOut[i*width_p + j] = 255; else imgBufOut[i*width_p + j] = t; } } }
测试:
#include"bmp.h" #include<iostream> using namespace std; int main() { char* fileName = "qianxun.bmp"; Bitmap* bmp = new Bitmap(); bmp->read(fileName); //bmp->translation(10, 10); //bmp->zoom(2, 2, 'n');//'n','l','c';默认为'n'邻近插值缩放 //bmp->rotate(90); bmp->fourier(); bmp->write("fourier.bmp"); delete bmp; return 1; }