一,颜色特征:
简单点来说就是将一幅图上的各个像素点颜色统计出来,适用颜色空间:RGB,HSV等颜色空间,
具体操作:量化颜色空间,每个单元(bin)由单元中心代表,统计落在量化单元上的像素数量
量化颜色直方图(HSV空间)
缺点:稀疏,量化问题
聚类颜色直方图:
适用颜色空间:Lab等颜色空间
操作:使用聚类算法对所有像素点颜色向量进行聚类
单元(bin)由聚类中心代表
解决稀疏问题
二,几何特征
边缘:像素明显变化的区域,含有丰富的语义信息
边缘定义:像素值快速变化的区域
边缘提取:
先噪声处理,高斯去噪,在使用一阶导数获取极值
高斯滤波一阶求导:
梯度变化最快方向:
三,基于特征点的特征描述子
不同的观测方式,物体的大小,形状,明暗会有不同,依然可以判断为同一物体
Harris角点(corner):
在任何方向上移动小观察窗,导致大的像素变动
代码:
def harris_corner():import numpy as npimport cv2filename = './data/chessboard.png'img = cv2.imread(filename)img=cv2.resize(img,(200,200))gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)gray = np.float32(gray)dst = cv2.cornerHarris(gray, 2, 3, 0.04)# result is dilated for marking the corners, not importantdst = cv2.dilate(dst, None)# Threshold for an optimal value, it may vary depending on the image.img[dst > 0.01 * dst.max()] = [0, 0, 255]cv2.imshow('dst', img)if cv2.waitKey(0) & 0xff == 27:cv2.destroyAllWindows()
打印结果:
(1)SIFT特征:基于尺度空间不变的特征,4×4网格,8方向直方图,总共128维特征向量
特点:具有良好的不变性,少数物体也能产生大量SIFT特征
代码:
def sift():import numpy as npimport cv2img = cv2.imread('./data/home.jpg')gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)sift = cv2.xfeatures2d.SIFT_create()kp = sift.detect(gray, None)img = cv2.drawKeypoints(gray, kp, img)cv2.imshow("SIFT", img)cv2.imwrite('sift_keypoints.jpg', img)cv2.waitKey(0)cv2.destroyAllWindows()
结果:
(2)LBP(局部二值模式):每个像素点与周围点大小比较,多个bit组成一个数,统计每个数的直方图,
LBP特征具有灰度不变性和旋转不变性等显著优点。
(3)SURF,为了保证旋转不变性,在SURF中,统计特征点领域内的Harr小波特征。
代码:
def surf():import numpy as npimport cv2img = cv2.imread('./data/butterfly.jpg', 0)surf = cv2.xfeatures2d.SURF_create(400)# kp, des = surf.detectAndCompute(img,None)surf.setHessianThreshold(50000)kp, des = surf.detectAndCompute(img, None)img2 = cv2.drawKeypoints(img, kp, None, (255, 0, 0), 4)cv2.imshow('surf', img2)cv2.waitKey(0)cv2.destroyAllWindows()
(4)ORB特征基于FAST角点的特征点检测
def orb():import numpy as npimport cv2 as cvimport matplotlib.pyplot as pltimg1 = cv.imread('./data/box.png', 0) # queryImageimg2 = cv.imread('./data/box_in_scene.png', 0) # trainImage# Initiate ORB detectororb = cv.ORB_create()# find the keypoints and descriptors with ORBkp1, des1 = orb.detectAndCompute(img1, None)kp2, des2 = orb.detectAndCompute(img2, None)# create BFMatcher objectbf = cv.BFMatcher(cv.NORM_HAMMING, crossCheck=True)# Match descriptors.matches = bf.match(des1, des2)# Sort them in the order of their distance.matches = sorted(matches, key=lambda x: x.distance)# Draw first 10 matches.img3 = cv.drawMatches(img1, kp1, img2, kp2, matches[:20], None, flags=2)plt.imshow(img3), plt.show()
(5)Gabor滤波:用于边缘提取的线性滤波器,三角函数+高斯函数=Gabor滤波器
基于sift拼接:Stitcher.py
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])# 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
def image_stich():from opencv.Stitcher import Stitcherimport cv2# 读取拼接图片imageA = cv2.imread("./data/left_01.png")imageB = cv2.imread("./data/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()