内容发布更新时间 : 2024/6/7 10:17:49星期一 下面是文章的全部内容请认真阅读。

PERFLAB实验报告

【实验目的】优化smooth函数。 【函数介绍】

Rotate：将图片逆时针旋转90°。 Smooth：用3*3的窗进行图片的均值滤波。

【程序优化】

Smooth： 1.1初始代码：

void naive_smooth(int dim, pixel *src, pixel *dst) {

int i, j;

for (i = 0; i < dim; i++) }

代码频繁调用avg函数，avg函数也频繁调研initialize_pixel_sum、sccumulate_sum、assign_sum_to_pixel函数。应该减少函数调用的时间开销。因此将所有函数都写在smooth内部，不再调用avg。

像素点分成图片四个角、图片四条边、图片内部三块分别进行处理。对角而言只需要4个像素点的均值，对于边而言为6个像素点均值，图片内部则需要9个像素点均值。

如图所示图片内部对于每行相邻的像素点A、B，滤波时其6个像素块重叠（绿色），对B而言，完全可以用上一步处理A后所得到的值，仅需加上B右边3个像素点（蓝色），并减去A多余的像素点（红色）即可。 每次所需要的数据由9个变为7个。

1.2改动代码：

void smooth(int dim, pixel *src, pixel *dst) {

dst[0].red=(src[0].red+src[1].red+src[dim].red+src[dim+1].red)>>2; //用\代替\节省时间 >2;

dst[dim*dim-1].red=(src[dim*dim-1].red+src[dim*dim-2].red+src[(dim-1)*dim-1].red+src[(dim-1)*dim-2].red)>>2; dst[dim*dim-1].blue=(src[dim*dim-1].blue+src[dim*dim-2].blue+src[(dim-1)*dim-1].blue+src[(dim-1)*dim-2].blue)>>2; dst[dim*dim-1].green=(src[dim*dim-1].green+src[dim*dim-2].green+src[(dim-1)*dim-1].green+src[(dim-1)*dim-2].green)>>dst[(dim-1)*dim].green=(src[(dim-1)*dim].green+src[(dim-1)*dim+1].green+src[(dim-2)*dim].green+src[(dim-2)*dim+1].green)>>2;

dst[(dim-1)*dim].red=(src[(dim-1)*dim].red+src[(dim-1)*dim+1].red+src[(dim-2)*dim].red+src[(dim-2)*dim+1].red)>>2; dst[(dim-1)*dim].blue=(src[(dim-1)*dim].blue+src[(dim-1)*dim+1].blue+src[(dim-2)*dim].blue+src[(dim-2)*dim+1].blue)>dst[dim-1].red=(src[dim-1].red+src[dim-2].red+src[dim*2-2].red+src[dim*2-1].red)>>2; dst[dim-1].blue=(src[dim-1].blue+src[dim-2].blue+src[dim*2-2].blue+src[dim*2-1].blue)>>2; dst[dim-1].green=(src[dim-1].green+src[dim-2].green+src[dim*2-2].green+src[dim*2-1].green)>>2; dst[0].blue=(src[0].blue+src[1].blue+src[dim].blue+src[dim+1].blue)>>2; dst[0].green=(src[0].green+src[1].green+src[dim].green+src[dim+1].green)>>2; int i,j,lastr,lastb,lastg; int row = dim; int curr;

for (j = 0; j < dim; j++)

dst[RIDX(i, j, dim)] = avg(dim, i, j, src);

2; //四个角

for (j=1; j < dim-1; j++) {

dst[j].red=(src[j].red+src[j-1].red+src[j+1].red+src[j+dim].red+src[j+1+dim].red+src[j-1+dim].red)/6;

dst[j].green=(src[j].green+src[j-1].green+src[j+1].green+src[j+dim].green+src[j+1+dim].green+src[j-1+dim].green)/6; dst[j].blue=(src[j].blue+src[j-1].blue+src[j+1].blue+src[j+dim].blue+src[j+1+dim].blue+src[j-1+dim].blue)/6; }

for (j=dim*(dim-1)+1; j < dim*dim-1; j++) {

dst[j].red=(src[j].red+src[j-1].red+src[j+1].red+src[j-dim].red+src[j+1-dim].red+src[j-1-dim].red)/6;

dst[j].green=(src[j].green+src[j-1].green+src[j+1].green+src[j-dim].green+src[j+1-dim].green+src[j-1-dim].green)/6; dst[j].blue=(src[j].blue+src[j-1].blue+src[j+1].blue+src[j-dim].blue+src[j+1-dim].blue+src[j-1-dim].blue)/6; }

for (i=dim; i < dim*(dim-1); i+=dim) {

dst[i].red=(src[i].red+src[i-dim].red+src[i+1].red+src[i+dim].red+src[i+1+dim].red+src[i-dim+1].red)/6;

dst[i].green=(src[i].green+src[i-dim].green+src[i+1].green+src[i+dim].green+src[i+1+dim].green+src[i-dim+1].green)/6; dst[i].blue=(src[i].blue+src[i-dim].blue+src[i+1].blue+src[i+dim].blue+src[i+1+dim].blue+src[i-dim+1].blue)/6; }

for (i=dim+dim-1; i < dim*dim-1; i+=dim) {

dst[i].red=(src[i].red+src[i-1].red+src[i-dim].red+src[i+dim].red+src[i-dim-1].red+src[i-1+dim].red)/6;

dst[i].green=(src[i].green+src[i-1].green+src[i-dim].green+src[i+dim].green+src[i-dim-1].green+src[i-1+dim].green)/6; dst[i].blue=(src[i].blue+src[i-1].blue+src[i-dim].blue+src[i+dim].blue+src[i-dim-1].blue+src[i-1+dim].blue)/6; } //四条边

lastg=src[row-dim].green+src[row-dim+1].green+src[row-dim+2].green+src[row].green+src[row+1].green+src[row+2].green+

dst[row+1].red=lastr/9; dst[row+1].blue=lastb/9;

dst[row+1].green=lastg/9; //内部每行第一个点的滤波结果，由于循环内部耗费时间会严重影响程序的耗时，因此对于lastr等的计算，for(j=2;j

curr=row+j;

src[row+dim].green+src[row+dim+1].green+src[row+dim+2].green;

lastb=src[row-dim].blue+src[row-dim+1].blue+src[row-dim+2].blue+src[row].blue+src[row+1].blue+src[row+2].blue+src[row+dim].blue+src[row+dim+1].blue+src[row+dim+2].blue;

lastr=src[row-dim].red+src[row-dim+1].red+src[row-dim+2].red+src[row].red+src[row+1].red+src[row+2].red+src[row+dim].red+src[row+dim+1].red+src[row+dim+2].red;

for(i=1;i

像素点的顺序都是计算最优的

lastr=lastr-src[curr-dim-2].red+src[curr-dim+1].red-src[curr-2].red+src[curr+1].red-src[curr+dim-2].red+src[curr+dim

lastb=lastb-src[curr-dim-2].blue+src[curr-dim+1].blue-src[curr-2].blue+src[curr+1].blue-src[curr+dim-2].blue+src[cur

lastg=lastg-src[curr-dim-2].green+src[curr-dim+1].green-src[curr-2].green+src[curr+1].green-src[curr+dim-2].green+sr

}

dst[curr].red=lastr/9; dst[curr].blue=lastb/9;

dst[curr].green=lastg/9; //内部其他点参考该行前一个点的像素值得到结果

+1].red;

r+dim+1].blue;

c[curr+dim+1].green;

}

row+=dim;

} //图片内部

2.3程序跑分：

2.4实验出现的问题：

Smooth刚开始在处理图片内部时，循环内部的代码是： 每行第一个像素点的滤波均值（这里以red为例）

dst[row+1].red=(src[row-dim].red+src[row-dim+1].red+src[row-dim+2].red+src[row].red+src[row+1].red+src[row+2].red+src[row+dim].red+src[row+dim+1].red+src[row+dim+2].red)/9，那么lastr=dst[row+1].red (即该点周围九个像素点rgb的均值)，则该行其余像素点都可以表示为dst[curr].red=lastr+(-src[curr-dim-2].red+src[curr-dim+1].red-src[curr-2].red+src[curr+1].red-src[curr+dim-2].red+src[curr+dim+1].red)/9，那么lastr=dst[curr].red。

运行后报错。观察发现所得rgb与理论的rgb相差不大，可能是因为计算过程中对数字的有效位数限制导致最后结果的差异。在代码中lastr为上一点计算rgb的均值，因此可能是在做除法过程中出现估算，导致相加后的差异。因此代码改动为现在的代码，即令lastr为上一点滤波所需像素点rgb的和。 对于每行第一个点：

lastr=src[row-dim].red+src[row-dim+1].red+src[row-dim+2].red+src[row].red+src[row+1].red+src[row+2].red+src[row+dim].red+src[row+dim+1].red+src[row+dim+2].red，那么dst[row+1].red=lastr/9。 对于该行其他像素点：

lastr=lastr-src[curr-dim-2].red+src[curr-dim+1].red-src[curr-2].red+src[curr+1].red-src[curr+dim-2].red+src[curr+dim+1].red，那么dst[curr].red=lastr/9。 运行后正确，说明猜测正确。