SIFT特征

?近日把SIFT特征提取及匹配算法推导了一遍，就開始我个人的一个认识吧。

?关于SIFT的算法介绍见最以下的链接。

?假设用SIFT的话。MATLAB的SIFT的code能够学习。个人觉得还是C写的SIFT比較好。

?VLFeat提供C写的SIFT特征提取算法代码，调用例如以下：

    //读入图像
    char *ImagePath="10011.jpg";  
	IplImage *Image=cvLoadImage(ImagePath,0);  
	//  int min=0;  
	//  min=Image->width>Image->height?Image->height:Image->width; 
	//sift特征提取參数设置
	int noctaves=4,nlevels=2,o_min=0;  
	// noctaves=(int)(log(min)/log(2));  
	//载入vlfeat库的sift特征库函数
	vl_sift_pix *ImageData=new vl_sift_pix[Image->height*Image->width];  
	unsigned char *Pixel;  
	for (int i=0;i<Image->height;i++)  
	{  
		for (int j=0;j<Image->width;j++)  
		{  
			Pixel=(unsigned char*)(Image->imageData+i*Image->width+j);  
			ImageData[i*Image->width+j]=*(Pixel);  
		}  
	}  
	VlSiftFilt *SiftFilt=NULL;  
	SiftFilt=vl_sift_new(Image->width,Image->height,noctaves,nlevels,o_min);  
	int KeyPoint=0;  
	int idx=0;  
	if (vl_sift_process_first_octave(SiftFilt,ImageData)!=VL_ERR_EOF)  
	{  
		while (true)  
		{  
			//计算每组中的关键点  
			vl_sift_detect(SiftFilt);  
			//遍历并绘制每一个点  
			KeyPoint+=SiftFilt->nkeys;  
			VlSiftKeypoint *pKeyPoint=SiftFilt->keys;  
			for (int i=0;i<SiftFilt->nkeys;i++)  
			{  
				VlSiftKeypoint TemptKeyPoint=*pKeyPoint;  
				pKeyPoint++;  
				cvDrawCircle(Image,cvPoint(TemptKeyPoint.x,TemptKeyPoint.y),TemptKeyPoint.sigma/2,CV_RGB(255,0,0));  
				idx++;  
				//计算并遍历每一个点的方向  
				double angles[4];  
				int angleCount=vl_sift_calc_keypoint_orientations(SiftFilt,angles,&TemptKeyPoint);  
				for (int j=0;j<angleCount;j++)  
				{  
					double TemptAngle=angles[j];  
					printf("%d: %f\n",j,TemptAngle);  
					//计算每一个方向的描写叙述  
					float *Descriptors=new float[128];  
					vl_sift_calc_keypoint_descriptor(SiftFilt,Descriptors,&TemptKeyPoint,TemptAngle);  
					int k=0;  
					//输出128维的值
					while (k<128)  
					{  
						printf("%d: %f",k,Descriptors[k]);  
						printf("; ");  
						k++;  
					}  

					printf("\n");  
					delete []Descriptors;  
					Descriptors=NULL;  
				}  
			}  
			//下一阶  
			if (vl_sift_process_next_octave(SiftFilt)==VL_ERR_EOF)  
			{  
				//
				break;  
			}  
			//free(pKeyPoint);  
			KeyPoint=NULL;  
		}  
	}  
	vl_sift_delete(SiftFilt);  
	delete []ImageData;  
	ImageData=NULL;  
	cvNamedWindow("Source Image",1);  
	cvShowImage("Source Image",Image);  

Mat image=imread("1.jpg");
	Mat iamgeGray=imread("1.jpg",0);
	Mat descriptors;
	vector<KeyPoint> keypoints;

	SiftFeatureDetector sift2(0.06f,10.0);
	sift2.detect(iamgeGray,keypoints);


	drawKeypoints(image,keypoints,image,Scalar(255,0,255));
	imshow("test",image);

	waitKey();
	return 0;

SIFT的改进型：

FAST-SIFT （Dom金字塔代替Dog金字塔） ?
PCA-SIFT（降低特征子匹配） ? ? ? ? ? ? ?

上面两种在改进型SIFT特征提取分析见兴许博文。。。。

SIFT特征提取分析相关链接：

http://blog.csdn.net/abcjennifer/article/details/7639681

https://github.com/robwhess/opensift/tree/master/src

http://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf

http://blog.csdn.net/abcjennifer/article/details/7365882

http://en.wikipedia.org/wiki/Scale-invariant_feature_transform#David_Lowe.27s_method

http://blog.sciencenet.cn/blog-613779-475881.html

http://www.cnblogs.com/linyunzju/archive/2011/06/14/2080950.html

http://www.cnblogs.com/linyunzju/archive/2011/06/14/2080951.html

http://blog.csdn.net/ijuliet/article/details/4640624

http://www.cnblogs.com/cfantaisie/archive/2011/06/14/2080917.html?