[Python] optimization of polyfit on 3D data set

I am trying to implement a non-uniformity correction algorithm on infrared images. Basically I have a set of images of a black body, which is supposed to be at a very precise and homogeneous temperature. But due to camera hardware (sensor, optics...), the image you get is not homogeneous.

I have images (640*512 pix) at different temperatures (around 10 images from 18°C to 30°C), and I need to correct the values of each pixel, to fit the mean value of the central area (supposed to be the correct value).

So, for each pixel, I need to fit a polynomial function p(order 5):

x=p(y)


where x is the value of the pixel, and y is the mean value of the central area.

For now I vectorized my images to avoid double looping, and loop over each pixel:

import numpy as np
interp_degree=5
nb_of_pixels=640*512
image =np.random.rand(640,512)
mean_value=np.arange(18,28,1)
a=image.flatten()
DL=np.transpose([a for i in range(10)]) # I simplified up to this point for the example

K=np.empty((nb_of_pixels,interp_degree+1))
for i in range(len(DL)): # loop over each pixels
K=np.polyfit(DL,mean_value,interp_degree) # fitting
DLC_NUC=np.polyval(K,DL) # eval the corrected value of the pixel for later use


This loop take about 200 sec on my computer, and I would like to reduce this time. polyfit and polyval already use multiprocessing, so this is not a way.

Is there ways to improve this?

Continue reading...