图像拼接算法研究

一、算法一

# SIFT的BF匹配

import numpy as np
import cv2

class Stitcher:

    # 拼接函数
    def stitch(self, images, ratio = 0.75, reprojThresh = 4.0, showMatches = False):
        # 读取图像
        imageB, imageA = images
        # 计算特征点和特征向量
        kpsA, featureA = self.detectAndDescribe(imageA)
        print(kpsA)
        print(featureA)
        print(kpsA.shape)
        print(featureA.shape)
        kpsB, featureB = self.detectAndDescribe(imageB)
        print(kpsB.shape)
        print(featureB.shape)
        # 匹配两张图片的特征点
        M = self.matchKeypoints(kpsA, kpsB, featureA, featureB, ratio, reprojThresh)

        # 没有匹配点，退出
        if not M:
            return None

        matches, H, status = M
        # 将图片A进行视角变换 中间结果（透视变换）
        result = cv2.warpPerspective(imageA, H, (imageA.shape[1] + imageB.shape[1], imageA.shape[0]))
        cv2.imshow('1', result)
        # 将图片B传入]
        result[0:imageB.shape[0], 0:imageB.shape[1]] = imageB
        self.cv_show('result', result)

        # 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 result


    def detectAndDescribe(self, image):
        # 转换为灰度图
        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

        # 建立SIFT生成器
        descriptor = cv2.xfeatures2d.SIFT_create()
        # 检测特征点并计算描述子
        # kps: 关键点信息，包括位置，尺度，方向信息
        # features: 关键点描述符，每个关键点对应128个梯度信息的特征向量
        kps, features = descriptor.detectAndCompute(gray, None)

        kps = np.float32([kp.pt for kp in kps])
        # 得到特征点和特征向量
        return kps, features

    def matchKeypoints(self, kpsA, kpsB, featureA, featureB, ratio, reprojThresh):
        # 建立暴力匹配器
        matcher = cv2.BFMatcher()

        # 使用KNN检测来自AB图的SIFT特征匹配
        rawMatches = matcher.knnMatch(featureA, featureB, 2)
        print(type(rawMatches))
        print(len(rawMatches))

        # 过滤
        matches = []
        for m in rawMatches:
            if len(m) == 2 and m[0].distance < m[1].distance * ratio:
                matches.append((m[0].trainIdx, m[0].queryIdx))
        print(matches)
        print(len(matches))

        if len(matches) > 4:
            # 获取匹配对的点坐标
            ptsA = np.float32([kpsA[i] for (_, i) in matches])
            print("ptsA的个数：", len(ptsA))
            ptsB = np.float32([kpsB[i] for (i, _) in matches])
            print(ptsB)
            # 计算H矩阵(计算单应性矩阵)（计算透视变换必须要4个点以上）
            H, status = cv2.findHomography(ptsA, ptsB, cv2.RANSAC, reprojThresh)
            print(H)

            return matches, H, status

    # 展示图像
    def cv_show(self,name, img):
        cv2.imshow(name, img)
        cv2.waitKey(0)
        cv2.destroyAllWindows()

    def 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] = imageA
        vis[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

# 读取图片
img1 = cv2.imread('1.jpg')
img1 = cv2.resize(img1, dsize=(1920,1080))
img2 = cv2.imread('2.jpg')
img2 = cv2.resize(img2, dsize=(1920,1080))

# 图片拼接
stitcher = Stitcher()
result, vis = stitcher.stitch([img1, img2], showMatches=True)

cv2.imwrite("1_1.jpg", result)
cv2.imwrite("2_1.jpg", vis)

# cv2.imshow('1', img1)
# cv2.imshow('2', img2)
# cv2.imshow('keypoints matches', vis)
# cv2.imshow('result', result)
# cv2.waitKey(0)
# cv2.destroyWindow()

二、算法二

1.下载vlfeat-0.9.20，Index of /download (vlfeat.org)

 找到vlfeat-0.9.20/bin/win64下的sift.exe

2.下载PVC包，找到PCV-master\PCV\localdescriptors\sift.py文件的18行，

cmmd = str("D:\\app\\vlfeat-0.9.20\\bin\\win64\\sift.exe "+imagename+" --output="+resultname+
            " "+params)

将vlfeat-0.9.20/bin/win64下的sift.exe的绝对路径添加到上面，注意.exe后有一个空格。

在代码的最后面调价运行

if __name__ == '__main__':
    imname = ('0.jpg')               #待处理图像路径
    im=Image.open(imname)
    process_image(imname, '0.sift')

这就就可以生成与图片对应的.sift文件

问题：调用sift.process_image片段时生成的sift特征文件大小为0字节

原因分析：vlfeat0.9.21版本太高导致部分电脑提取特征为空

解决办法：更换版本为vlfeat0.9.20版本，部分电脑更换版本后可能会出现vcomp100.dll文件找不到问题，可下载对应的dll文件，下载链接https://cn.dll-files.com/vcomp100.dll.html，64位系统对应64位，下载后放入C:\Windows\SysWOW64和C:\Windows\System32，有则替换，亲测可运行成功。

3.利用图像与生成的.sift文件进行三张图片的拼接（修改代码可以实现任意多张图片的拼接）

# -*- coding: utf-8 -*-
from pylab import *
from numpy import *
from PIL import Image
# If you have PCV installed, these imports should work
from PCV.geometry import homography, warp
from PCV.localdescriptors import sift
"""
This is the panorama example from section 3.3.
"""
# set paths to data folder
featname = ['D:\\PCV-master\\全景拼接4\\' + str(i + 1) + '.sift' for i in range(3)]
imname = ['D:\\PCV-master\\全景拼接4\\' + str(i + 1) + '.jpg' for i in range(3)]

# extract features and match
l = {}
d = {}
for i in range(3):
    sift.process_image(imname[i],featname[i])
    l[i],d[i] = sift.read_features_from_file(featname[i])
matches = {}
for i in range(2):
    matches[i] = sift.match(d[i+1],d[i])
    print(matches)
# visualize the matches (Figure 3-11 in the book)
for i in range(2):
    im1 = array(Image.open(imname[i]))
    im2 = array(Image.open(imname[i+1]))
    # figure()
    # sift.plot_matches(im2,im1,l[i+1],l[i],matches[i],show_below=True)
# function to convert the matches to hom. points
def convert_points(j):
    ndx = matches[j].nonzero()[0]
    fp = homography.make_homog(l[j+1][ndx,:2].T)
    ndx2 = [int(matches[j][i]) for i in ndx]
    tp = homography.make_homog(l[j][ndx2,:2].T)
    # switch x and y - TODO this should move elsewhere
    fp = vstack([fp[1],fp[0],fp[2]])
    tp = vstack([tp[1],tp[0],tp[2]])
    return fp,tp
# estimate the homographies
model = homography.RansacModel()
fp,tp = convert_points(0)
H_01 = homography.H_from_ransac(fp,tp,model)[0] #im 0 to 1
tp,fp = convert_points(1) #NB: reverse order
H_21 = homography.H_from_ransac(fp,tp,model)[0] #im 2 to 1
# warp the images
delta = 2000 # for padding and translation
im1 = array(Image.open(imname[0]), "uint8")
im2 = array(Image.open(imname[1]), "uint8")
im_01 = warp.panorama(H_01,im1,im2,delta,delta)
im1 = array(Image.open(imname[2]), "f")
im_21 = warp.panorama(H_21,im1,im_01,delta,delta)
figure()
imshow(array(im_21, "uint8"))
axis('off')
savefig("result.png",dpi=300)
show()