这段时间在用OPENCV2实现图像运动模糊图像复原功能，其中需要合成不同模糊长度和模糊角度的运动模糊图像，由于没有直接可用的函数，于是自己编了一个，采用OPENCV2，代码如下：
<pre class="cpp" name="code">void genaratePsf(Mat &psf, double len,double angle){double half=len/2;double alpha = (angle-floor(angle/ 180) *180) /180* PI;double cosalpha = cos(alpha);double sinalpha = sin(alpha);int xsign;if (cosalpha < 0){xsign = -1;}else{if (angle == 90){xsign = 0;}else{xsign = 1;}}int psfwdt = 1;int sx = (int)fabs(half*cosalpha + psfwdt*xsign – len*EPS);int sy = (int)fabs(half*sinalpha + psfwdt – len*EPS);Mat_<double> psf1(sy, sx, CV_64F);Mat_<double> psf2(sy * 2, sx * 2, CV_64F);int row = 2 * sy;int col = 2 * sx;/*为减小运算量，先计算一半大小的PSF*/for (int i = 0; i < sy; i++){double* pvalue = psf1.ptr<double>(i);for (int j = 0; j < sx; j++){pvalue[j] = i*fabs(cosalpha) – j*sinalpha;double rad = sqrt(i*i + j*j);if (rad >= half && fabs(pvalue[j]) <= psfwdt){double temp = half – fabs((j + pvalue[j] * sinalpha) / cosalpha);pvalue[j] = sqrt(pvalue[j] * pvalue[j] + temp*temp);}pvalue[j] = psfwdt + EPS – fabs(pvalue[j]);if (pvalue[j] < 0){pvalue[j] = 0;}}}/*将模糊核矩阵扩展至实际大小*/for (int i = 0; i < sy; i++){double* pvalue1 = psf1.ptr<double>(i);double* pvalue2 = psf2.ptr<double>(i);for (int j = 0; j < sx; j++){pvalue2[j] = pvalue1[j];}}for (int i = 0; i < sy; i++){for (int j = 0; j < sx; j++){psf2[2 * sy -1 – i][2 * sx -1 – j] = psf1[i][j];psf2[sy + i][j] = 0;psf2[i][sx + j] = 0;}}/*保持图像总能量不变，归一化矩阵*/double sum = 0;for (int i = 0; i < row; i++){for (int j = 0; j < col; j++){sum+= psf2[i][j];}}psf2 = psf2 / sum;if (cosalpha>0){flip(psf2, psf2, 0);}//cout << "psf2=" << psf2 << endl;psf = psf2;
模糊长度为20，模糊角度为40的模糊核，，效果如下图：
业精于勤，荒于嬉；行成于思，毁于随。
