scikit-image Gabor filter error: `filter weights array has incorrect shape`

Question

输入是灰度图像，转换为 130x130 numpy 矩阵。我总是收到错误消息：

Traceback (most recent call last):
  File "test_final.py", line 87, in <module>
    a._populate_gabor()

  File "C:\Users\Bears\Dropbox\School\Data Science\final.py", line 172, in _populate_gabor
    self.gabor_imgs[i] = self._matrix_2_1d(self._gabor_this(self.grey_imgs[i]),kernels[0])

  File "C:\Users\Bears\Dropbox\School\Data Science\final.py", line 179, in _gabor_this
    filtered = ndi.convolve(image, kernel, mode='reflect')

  File "C:\Users\Bears\Anaconda3\lib\site-packages\scipy\ndimage\filters.py", line 696, in convolve
    origin, True)

  File "C:\Users\Bears\Anaconda3\lib\site-packages\scipy\ndimage\filters.py", line 530, in _correlate_or_convolve
    raise RuntimeError('filter weights array has incorrect shape.')
RuntimeError: filter weights array has incorrect shape.

我的代码如下

def _populate_gabor(self):
    kernels = []
    for theta in range(self.gabor_range[0],self.gabor_range[1]):
        theta = theta / 4. * np.pi
        for sigma in (1, 3):
            for frequency in (0.05, 0.25):
                kernel = np.real(gabor_kernel(frequency, theta=theta,
                                      sigma_x=sigma, sigma_y=sigma))
                kernels.append(kernel)
    print (len(kernels))

    for i in range(self.length):
        self.gabor_imgs[i] = self._matrix_2_1d(self._gabor_this(self.grey_imgs[i]),kernels[0])


def _gabor_this(image, kernels): 
    feats = np.zeros((len(kernels), 2), dtype=np.double)
    for k, kernel in enumerate(kernels):
        filtered = ndi.convolve(image, kernel, mode='reflect')
        feats[k, 0] = filtered.mean()
        feats[k, 1] = filtered.var()
    return feats

我直接从 http://scikit-image.org/docs/dev/auto_examples/plot_gabor.html 的示例中获取了这段代码，但我不知道如何解决这个错误。任何帮助，将不胜感激。请注意，所有其他功能都与其他过滤器一起使用，而不是 gabor。

Answer 1

您好像在使用 scipy 中的 'ndimage.convolve' 函数。请记住，ndimage 提供了 "N" 维度卷积。所以如果你想让卷积起作用，图像和核必须有相同的维数。其中任何一个尺寸不正确都会导致您描述的错误。

根据您上面的评论，内核 (4,4,7) 无法与图像 (130,130) 进行卷积。尝试在卷积之前添加一个单一维度，然后在之后删除它。

img = np.zeros(shape=(130,130),dtype=np.float32)
img = img[:,:,None] # Add singleton dimension
result = convolve(img,kernel)
finalOutput = result.squeeze() # Remove singleton dimension

Answer 2

这是一个关于如何使用 ndimage.convole 对 RGB 图像（3 通道图像）应用滤镜的示例):

import cv2
import numpy as np
from PIL import Image
from pylab import plt
from scipy import ndimage

# Partly based on 
# Load an rgb image
img_in = np.array(Image.open("a_256x256_rgb_image.bmp")).reshape(256, 256, 3)

rgb_kernel = np.ones((5, 5, 3))

# A highpass filter
single_chan_kernel = np.array(
    [
        [-1, -1, -1, -1, -1],  # nopep8
        [-1, 1, 2, 1, -1],  # nopep8
        [-1, 2, 4, 2, -1],  # nopep8
        [-1, 1, 2, 1, -1],  # nopep8
        [-1, -1, -1, -1, -1],
    ]  # nopep8
)  # nopep8

rgb_kernel[:, :, 0] = single_chan_kernel
rgb_kernel[:, :, 1] = single_chan_kernel
rgb_kernel[:, :, 2] = single_chan_kernel

filtered_img = ndimage.convolve(img_in, rgb_kernel)

im = plt.imshow(filtered_img)
plt.show()

scikit-image Gabor filter error: `filter weights array has incorrect shape`

scikit-image Gabor filter error: `filter weights array has incorrect shape`

python

arrays

numpy

scikit-image