图像处理之Hessian矩阵提取关键点
程序员文章站
2022-07-15 22:34:28
...
一:大致的算法流程
1. 对每个像素点计算图像在X方向Y方向的二阶偏导数,计算图像的XY方向的导数
2. 根据第一步的计算结果,有Hessian Matrix计算D(h) = Ixx*Iyy - Ixy*Ixy
其中Ixx表示X方向的二阶偏导数
Iyy表示Y方向的二阶偏导数
Ixy表XY方向的二阶导数
3. 根据第二步计算出来的值使用3×3窗口实现非最大信号压制,
我的做法, 直接给了threshold值,这个其实不很对,真的懒,不想弄啦!
二:导数计算实现
关于一阶与二阶高斯偏导数计算请看这里:
http://blog.csdn.net/jia20003/article/details/16369143
三:程序效果
四:算法代码
package com.gloomyfish.image.harris.corner;
import java.awt.image.BufferedImage;
import java.util.ArrayList;
import java.util.List;
import com.gloomyfish.filter.study.AbstractBufferedImageOp;
public class HessianFeatureDetector extends AbstractBufferedImageOp {
private GaussianDerivativeFilter gdFilter;
private double minRejectThreshold = 4.1; // (r+1)^2/r
private List<HessianMatrix> pixelMatrixList;
public HessianFeatureDetector()
{
gdFilter = new GaussianDerivativeFilter();
pixelMatrixList = new ArrayList<HessianMatrix>();
}
@Override
public BufferedImage filter(BufferedImage src, BufferedImage dest) {
int width = src.getWidth();
int height = src.getHeight();
initSettings(height, width);
if ( dest == null )
dest = createCompatibleDestImage( src, null );
int[] inPixels = new int[width*height];
gdFilter.setDirectionType(GaussianDerivativeFilter.XX_DIRECTION);
BufferedImage bixx = gdFilter.filter(src, null);
getRGB( bixx, 0, 0, width, height, inPixels );
extractPixelData(inPixels, GaussianDerivativeFilter.XX_DIRECTION, height, width);
// YY Direction
gdFilter.setDirectionType(GaussianDerivativeFilter.YY_DIRECTION);
BufferedImage biyy = gdFilter.filter(src, null);
getRGB( biyy, 0, 0, width, height, inPixels );
extractPixelData(inPixels, GaussianDerivativeFilter.YY_DIRECTION, height, width);
// XY Direction
gdFilter.setDirectionType(GaussianDerivativeFilter.XY_DIRECTION);
BufferedImage bixy = gdFilter.filter(src, null);
getRGB( bixy, 0, 0, width, height, inPixels );
extractPixelData(inPixels, GaussianDerivativeFilter.XY_DIRECTION, height, width);
int[] outPixels = new int[width*height];
int index = 0;
for(int row=0; row<height; row++) {
int ta = 0, tr = 0, tg = 0, tb = 0;
for(int col=0; col<width; col++) {
index = row * width + col;
ta = 255;
HessianMatrix hm = pixelMatrixList.get(index);
double[] t = hm.getThreshold();
if(t[0] > minRejectThreshold)
{
tr = 127;
}
else
{
tr = 0;
}
if(t[1] > minRejectThreshold)
{
tg = 127;
}
else
{
tg = 0;
}
if(t[2] > minRejectThreshold)
{
tb = 127;
}
else
{
tb = 0;
}
outPixels[index] = (ta << 24) | (tr << 16) | (tg << 8) | tb;
}
}
setRGB( dest, 0, 0, width, height, outPixels );
return dest;
}
private void initSettings(int height, int width)
{
int index = 0;
for(int row=0; row<height; row++) {
for(int col=0; col<width; col++) {
index = row * width + col;
HessianMatrix matrix = new HessianMatrix();
pixelMatrixList.add(index, matrix);
}
}
}
private void extractPixelData(int[] pixels, int type, int height, int width)
{
int index = 0;
for(int row=0; row<height; row++) {
int ta = 0, tr = 0, tg = 0, tb = 0;
for(int col=0; col<width; col++) {
index = row * width + col;
ta = (pixels[index] >> 24) & 0xff;
tr = (pixels[index] >> 16) & 0xff;
tg = (pixels[index] >> 8) & 0xff;
tb = pixels[index] & 0xff;
HessianMatrix matrix = pixelMatrixList.get(index);
if(type == GaussianDerivativeFilter.XX_DIRECTION)
{
matrix.setXx(new double[]{tr, tg, tb});
}
if(type == GaussianDerivativeFilter.YY_DIRECTION)
{
matrix.setYy(new double[]{tr, tg, tb});
}
if(type == GaussianDerivativeFilter.XY_DIRECTION)
{
matrix.setXy(new double[]{tr, tg, tb});
}
}
}
}
}
转载请注明!
上一篇: 图像处理之应用卷积– 轧花与边缘检测
下一篇: 两个数组的交集2
推荐阅读