2023.8.19
为了在暑假内实现深度学习的进阶学习,特意学习一下传统算法,分享学习心得,记录学习日常
SIFT的百科:
SIFT = Scale Invariant Feature Transform, 尺度不变特征转换
全网最详细SIFT算法原理实现_ssift算法_Tc.小浩的博客-CSDN博客
在环境配置中要配置opencv:
pip install opencv-contrib-python
SIFT算法的三个计算步骤:
1,在DOG尺度空间中获取特征点;
2,关键点的方向估计(寻找主方向)
3,通过各关键点的特征向量(关键点的描述子生成)
进行两两比较找出相互匹配的若干对特征点,建立两图间景物间的对应关系,可以基于SIFT实现图像拼接
Code of SIFT and lena:
注意你是否有lena.png图像
import cv2 as cvimg = cv.imread('lena.png')
gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
sift = cv.xfeatures2d.SIFT_create()
# sift = cv.SIFT_create()
kp = sift.detect(gray, None)
img = cv.drawKeypoints(gray, kp, img)
cv.imshow("SIFT", img)
cv.imwrite('sift_keypoints.jpg', img)
cv.waitKey(0)
cv.destroyAllWindows()
Result is shown in these figrues : SIFT 提取了lena的特征点
基于SIFT的图片实现图片拼接:
代码是Copy大神的,注意有两个代码,运行第二个喔。代码所用的图片也附上!
import numpy as np
import cv2class Stitcher:# 拼接函数def stitch(self, images, ratio=0.75, reprojThresh=4.0, showMatches=False):# 获取输入图片(imageB, imageA) = images# 检测A、B图片的SIFT关键特征点,并计算特征描述子(kpsA, featuresA) = self.detectAndDescribe(imageA)(kpsB, featuresB) = self.detectAndDescribe(imageB)# 匹配两张图片的所有特征点,返回匹配结果M = self.matchKeypoints(kpsA, kpsB, featuresA, featuresB, ratio, reprojThresh)# 如果返回结果为空,没有匹配成功的特征点,退出算法if M is None:return None# 否则,提取匹配结果# H是3x3视角变换矩阵(matches, H, status) = M# 将图片A进行视角变换,result是变换后图片result = cv2.warpPerspective(imageA, H, (imageA.shape[1] + imageB.shape[1], imageA.shape[0]))# 将图片B传入result图片最左端result[0:imageB.shape[0], 0:imageB.shape[1]] = imageB# 检测是否需要显示图片匹配if showMatches:# 生成匹配图片vis = self.drawMatches(imageA, imageB, kpsA, kpsB, matches, status)# 返回结果return (result, vis)# 返回匹配结果return resultdef detectAndDescribe(self, image):# 将彩色图片转换成灰度图gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)# 建立SIFT生成器descriptor = cv2.xfeatures2d.SIFT_create()# 检测SIFT特征点,并计算描述子(kps, features) = descriptor.detectAndCompute(image, None)# 将结果转换成NumPy数组kps = np.float32([kp.pt for kp in kps])print(kps)# 返回特征点集,及对应的描述特征return (kps, features)def matchKeypoints(self, kpsA, kpsB, featuresA, featuresB, ratio, reprojThresh):# 建立暴力匹配器matcher = cv2.DescriptorMatcher_create("BruteForce")# 使用KNN检测来自A、B图的SIFT特征匹配对,K=2rawMatches = matcher.knnMatch(featuresA, featuresB, 2)matches = []for m in rawMatches:# 当最近距离跟次近距离的比值小于ratio值时,保留此匹配对if len(m) == 2 and m[0].distance < m[1].distance * ratio:# 存储两个点在featuresA, featuresB中的索引值matches.append((m[0].trainIdx, m[0].queryIdx))# 当筛选后的匹配对大于4时,计算视角变换矩阵if len(matches) > 4:# 获取匹配对的点坐标ptsA = np.float32([kpsA[i] for (_, i) in matches])ptsB = np.float32([kpsB[i] for (i, _) in matches])# 计算视角变换矩阵(H, status) = cv2.findHomography(ptsA, ptsB, cv2.RANSAC, reprojThresh)# 返回结果return (matches, H, status)# 如果匹配对小于4时,返回Nonereturn Nonedef drawMatches(self, imageA, imageB, kpsA, kpsB, matches, status):# 初始化可视化图片,将A、B图左右连接到一起(hA, wA) = imageA.shape[:2](hB, wB) = imageB.shape[:2]vis = np.zeros((max(hA, hB), wA + wB, 3), dtype="uint8")vis[0:hA, 0:wA] = imageAvis[0:hB, wA:] = imageB# 联合遍历,画出匹配对for ((trainIdx, queryIdx), s) in zip(matches, status):# 当点对匹配成功时,画到可视化图上if s == 1:# 画出匹配对ptA = (int(kpsA[queryIdx][0]), int(kpsA[queryIdx][1]))ptB = (int(kpsB[trainIdx][0]) + wA, int(kpsB[trainIdx][1]))cv2.line(vis, ptA, ptB, (0, 255, 0), 1)# 返回可视化结果return vis
from Stitcher import Stitcher
import cv2# 读取拼接图片
imageA = cv2.imread("image/left_01.png")
imageB = cv2.imread("image/right_01.png")# 把图片拼接成全景图
stitcher = Stitcher()
(result, vis) = stitcher.stitch([imageA, imageB], showMatches=True)# 显示所有图片
cv2.imshow("Image A", imageA)
cv2.imshow("Image B", imageB)
cv2.imshow("Keypoint Matches", vis)
cv2.imshow("Result", result)
cv2.waitKey(0)
cv2.destroyAllWindows()
代码使用的图片
效果是这样: