OpenCV image Stiching

Posted February 21, 2019 by Rokas Balsys



OpenCV job application task #part 1

Welcome to my third tutorial with OpenCV and python.

This tutorial will be different from previous two tutorials. So main idea why I am doing this is that some time ago I was applying for a job and received a task to do. So here is the requirements for a task: You have two similar photos JPEG(RGB) and TIFF(Grayscale) or pack of photos in two different folders. What I need to do:

• Find overlapping region betweet these two photos and crop RGB photo up to overlapping region.
• Save TIFF photo to 4th RGB image channel.
• Calculate stitching duration.
• Use multiprocessing to faster things up.
• Use script from CMD line.
• Save stitched images somewhere.

For this job I have two folders of images, but in this tutorial part I will use only two images below. In our next tutorial I will give you a link where you can download all these images if you like to test code by your self. So here is two images:


RGB.jpeg


NIR.jpeg

I won't go into details what I changed and why, if you want more details you can watch my video tutorial. Mainly what I did I copied parts of my code from my first and second OpenCV tutorials and put everything into "MainFunction". Then I created if __name__ == "__main__": new function, which always starts when we run our script and made that we call our MainFunction from there with image names. At the same time I measured time duration before doing image stitching and after, then I calculated how long it took.

You can download images from these links: RGB and NIR.

Our stitched image result:


_.png

Here is full tutorial code:

import cv2
import numpy as np
import time

resizer = 1
def MainFunction(JPEGimage, TIFFimage, number=''):
    img_ = cv2.imread(JPEGimage)
    img_ = cv2.resize(img_, (0,0), fx=resizer, fy=resizer)
    img1 = cv2.cvtColor(img_,cv2.COLOR_BGR2GRAY)

    img = cv2.imread(TIFFimage)
    img = cv2.resize(img, (0,0), fx=resizer, fy=resizer)
    img2 = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

    sift = cv2.xfeatures2d.SIFT_create()
    # find the key points and descriptors with SIFT
    kp1, des1 = sift.detectAndCompute(img1,None)
    kp2, des2 = sift.detectAndCompute(img2,None)
    #cv2.imshow('original_image_left_keypoints',cv2.drawKeypoints(img_,kp1,None))

    match = cv2.BFMatcher()
    matches = match.knnMatch(des1,des2,k=2)

    good = []
    for m,n in matches:
        if m.distance < 0.5*n.distance:
            good.append(m)
            
    if len(good) > 10:
        src_pts = np.float32([ kp1[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
        dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good ]).reshape(-1,1,2)

        M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)

    '''
    draw_params = dict(matchColor = (0,255,0), # draw matches in green color
                       singlePointColor = None,
                       flags = 2)

    img3 = cv2.drawMatches(img_,kp1,img,kp2,good,None,**draw_params)
    cv2.imshow("original_image_drawMatches.jpg", img3)
    '''
    warped_image = cv2.warpPerspective(img_,M,(img.shape[1], img.shape[0]))

    RGBA_image = cv2.cvtColor(warped_image, cv2.COLOR_RGB2RGBA)
    RGBA_image[:,:,3] = img2
    cv2.imwrite("_"+str(number)+".png",RGBA_image)

    
if __name__ == "__main__":
    start_time = time.time()

    MainFunction('RGB.jpeg', 'NIR.tiff')

    print("Time took:", time.time() - start_time)
    print("Finished")


In this short tutorial I wasn't explaining everything in details while writing this tutorial. If you want more details check my two previous tutorials and my video tutorials. So what is left for us to do is to write a code, that we could stitch single images and images from folders, do everything in multiprocessing and start process from cmd line.