Image Enhancement

scikit-shape includes algorithms for image enhancement that can be used for image preprocessing to improve performance of image analysis algorithms. Some of these enhancement algorithms act like denoising algorithms. They reduce noise or variation in the image while preserving or improving salience of important features, such as edges, in the image. These algorithms can be found in the module skshape.image.enhancement. Below we include an example, in which we apply a spatially-varying weighted smoothing to the given image to remove noise and clutter within regions.

Weighted image smoothing

The function weighted_smoothing() in module skshape.image.enhancement module can be used to apply a spatially-varying weighted smoothing to the given image. The weighting of the smoothing is such that it is strong within low-variation image regions, away from region boundaries, and it is minimal on edges or boundaries. Thus, it smoothes the image within regions, but keeps the edges sharp. The spatial weight of the smoothing is given by the edge indicator function, whose value is close to 1 in smooth areas of the image, but small on locations of high variation or high image gradient, such as edges. The definition of the edge indicator function g(x) is given by

where x is the spatial location on the image and I(x) is the image intensity function.

The weighted smoothing procedure can be applied to an image as shown below

>>> import matplotlib.pyplot as plt
>>> from skshape.image.enhancement import weighted_smoothing
>>> image = plt.imread('data/bacteria.jpg') / 255.0
>>> smoothed_image = weighted_smoothing( image )
>>> plt.figure()
>>> plt.subplot(1,2,1);  plt.axis('off');  plt.gray()
>>> plt.title('Original image');  plt.imshow(image)
>>> plt.subplot(1,2,2);  plt.axis('off');  plt.gray()
>>> plt.title('Smoothed image');  plt.imshow(smoothed_image)
>>> plt.show()

The amount of smoothing can be adjusted with the parameters diffusion_weight and data_weight of the function weighted_smoothing(). Increasing diffusion_weight will make the result smoother, and increasing data_weight will keep the result closer to the original image. The following shows the input image used by the example above, and the resulting smoothed image.