OpenCV PCA Compute in Python

Question

I'm loading a set of test images via OpenCV (in Python) which are 128x128 in size, reshape them into vectors (1, 128x128) and put them all together in a matrix to calculate PCA. I'm using the new cv2 libaries...

The code:

import os
import cv2 as cv
import numpy as np

matrix_test = None
for image in os.listdir('path_to_dir'):
    imgraw = cv.imread(os.path.join('path_to_dir', image), 0)
    imgvector = imgraw.reshape(128*128)
    try:
        matrix_test = np.vstack((matrix_test, imgvector))
    except:
        matrix_test = imgvector

# PCA
mean, eigenvectors = cv.PCACompute(matrix_test, np.mean(matrix_test, axis=0))

And it allways fails on the PCA part (I tested the image loading and all, the resulting matrix is how it should be)...the error I get is:

File "main.py", line 22, in

mean, eigenvectors = cv.PCACompute(matrix_test, np.mean(matri_test, axis=0))

cv2.error: /path/to/OpenCV-2.3.1/modules/core/src/matmul.cpp:2781: error: (-215) _mean.size() == mean_sz in function operator()

score 10 · Accepted Answer · answered Dec 19 '11 at 21:59

10

I think the problem is with the size of

np.mean(matrix_test, axis=0)

Its size is (128x128,) and not (1, 128x128). Thus the code below should work

mean, eigenvectors = cv.PCACompute(matrix_test, np.mean(matrix_test, axis=0).reshape(1,-1))

answered Dec 19 '11 at 21:59

Dat Chu

10,822
13
58
82

That did the trick...totally missed that, such a dumb mistake! Thanks! – Veles Dec 19 '11 at 23:44

score 7 · Answer 2 · answered Mar 24 '16 at 15:39

7

You can also put

cv.PCACompute(matrix_test, mean = np.array([]))

and the function computes the mean.

answered Mar 24 '16 at 15:39

Tomas

77
1
2

2

Good answer +1 this helped me a lot. I also found that `mean=None`, is equivalent to this solution. Cheers – DarkCygnus Dec 05 '17 at 19:08

OpenCV PCA Compute in Python

2 Answers2

Linked