Machine Learning Algorithms

Image convolution

Even if we're not building a complete deep learning model, we can test how convolution works with a simple example. The input image we're using is already provided by SciPy:

from scipy.misc import face>>> img = face(gray=True)

The original picture is shown here:

We're going to apply a Laplacian filter, which emphasizes the boundary of each shape:

import numpy as np>>> kernel = np.array(>>>    [[0, 1, 0],>>>     [1, -4, 0],>>>     [0, 1, 0]], >>>    dtype=np.float32)>>> cfilter = np.zeros((3, 3, 1, 1), dtype=np.float32)>>> cfilter[:, :, 0, 0] = kernel

The kernel must be repeated twice because the TensorFlow convolution function tf.nn.conv2d ...