Converting opencv remap code from c++ to python

Question

I am trying to convert c++ opencv cv2.remap code to python. I am not getting any error but result is not as expected.I am getting zoomed image

c++ code

int main()
{
    Mat img = imread("captcha1.jpg");
    float phase = -0.8 * CV_PI;
    float omega = 2.0 * CV_PI / img.cols;
    float amp = 15;
    Mat_<Vec2f> proj(img.size());
    for (int y=0; y<img.rows; y++) {
        for (int x=0; x<img.cols; x++) {
            float u = 0;
            float v = sin(phase + float(x) * omega) * amp;
            proj(y,x) = Vec2f(float(x) + u, float(y) + v);
        }
    }
    Mat corr;
    cv::remap(img, corr, proj, cv::Mat(), INTER_LINEAR);
    imshow("in",img);
    imshow("out",corr);
    waitKey();
}

python code :

from __future__ import print_function
import cv2
import numpy as np
from past.builtins import xrange
def update():
   for j in xrange(rows):
        for i in xrange(cols):

            phase = -0.8 * np.pi
            omega = 2.0 * np.pi / cols
            amp = 15
            u = -100
            v = np.sin(phase + float(i) * omega) * amp
            value_xy = (float(i) + u) + (float(j) + v)
            map_x.itemset((j, i), value_xy)
            map_y.itemset((j, i), 0)

img = cv2.imread('test.jpg')
map_x = np.zeros(img.size, np.float32)
map_y = np.zeros(img.shape[:2], np.float32)
rows, cols = img.shape[:2]
update()
dst = cv2.remap(img, map_x, map_y, cv2.INTER_LINEAR)
cv2.imshow('dst', dst)
cv2.waitKey(0)
cv2.destroyAllWindows()

Original image :

C++ Result :

Python result:

Bid

Why u = 0 in c++ and u = -100 in Python? Why not pass rows, cols to update as parameters, rather then relying on global variables? IMHO map_y can just be None. – Davidadavidde

I also don't understand why you make your life harder by changing variable names and the structure of the code when there's absolutely no reason for it. I mean, look at this -- isn't that easier to compare side-by-side with the original C++ implementation? – Davidadavidde

Once you have that, you can try to optimize things. e.g. get rid of those nested for loops, and replace them with vectorized numpy operations. Say, like this – Davidadavidde

@DanMašek : Thanks for the help, confused on this # NB: Mat_<Vec2f> -> 2 channels proj = np.zeros((img.shape[0], img.shape[1], 2), np.float32) – Bid

Answer 1

This is my result:

Generaly, I prefer the the vectorized implementation to the for-loop implementation in Python. Here is my code:

#!/usr/bin/python3
# 2018.09.23 12:24 (CST)
import cv2 
import numpy as np 

fname = "remap.jpg"
img = cv2.imread(fname)
nh, nw = img.shape[:2]

PI = 3.141592653589793
phase = -0.8 * PI
omega = 2.0 * PI / nw
amp = 15

xs, ys = np.meshgrid(np.arange(0, nw), np.arange(0, nh))
ys = np.sin(phase+xs*omega)*amp + ys
xs = np.float32(xs)
ys = np.float32(ys)

dst= cv2.remap(img, xs, ys, cv2.INTER_CUBIC)
cv2.imwrite("dst.png", dst)

Handle