Numpy dot product

Question

I have an image represented with a Numpy array, i.e. each pixel is an array [r,g,b]. Now, I want to convert it in YUV using matrix multiplication and trying not to use loops.

self.yuv=self.rgb
self.yuv=dot([[   0.299,  0.587,    0.114  ],
              [-0.14713, -0.28886,  0.436  ],
              [   0.615, -0.51499, -0.10001]], 
             self.yuv[:,:])

I get the error - objects not aligned. I guess that's because self.yuv[i,j] is not a vertical vector. transpose doesn't help.

Any ideas?

Isn't rgb a 3 dimensional array. Can you tell us what gives `self.rgb.ndim`? — petrichor, May 28 '13 at 18:22
So, you shouldn't use `self.yuv[:,:]`. Here's a good answer for HSV conversion. You may like to check it: http://stackoverflow.com/questions/4890373/detecting-thresholds-in-hsv-color-space-from-rgb-using-python-pil/4890878#4890878 — petrichor, May 28 '13 at 18:28

score 5 · Accepted Answer · answered May 28 '13 at 20:09

Your matrix has shape (3, 3) while your image has shape (rows, cols, 3) and np.dot does "a sum product over the last axis of a and the second-to-last of b."

The simplest solution is to reverse the order of the operands inside np.dot and transpose your conversion matrix:

rgb2yuv = np.array([[0.299, 0.587, 0.114],
                    [-0.14713, -0.28886, 0.436],
                    [0.615, -0.51499, -0.10001]])
self.yuv = np.dot(self.rgb, rgb2yuv.T)

Numpy dot product

1 Answers1

Linked