Image Restoration and Reconstruction with Inpainting

Image Restoration

• Image restoration is an image processing field in Python
• It involves restoring damaged images with techniques such as inpainting

Image Reconstruction

• Fixing damaged images and removing unwanted elements like text, logos, or objects are aspects of image reconstruction

Inpainting

• Inpainting reconstructs lost parts, focusing on non-damaged regions of images
• The damaged pixels are set as a mask

Mask images

• A mask image is one where some pixel intensity values are zero, and others are non-zero (damaged)
• Damaged regions (white pixels) are used, with damaged areas of an image serving as a mask

Image reconstruction in scikit-image

• The process involves obtaining the mask of a defective image using get_mask
• Then applying inpainting to this damaged image uses the mask via the inpaint.inpaint_biharmonic with the defect_image, and mask and channel_axis=-1) parameters
• Displaying the resulting restored image is done through show_image

Noise in Digital Images

• Noise in digital images refers to random variations of brightness or color information
• Noise can manifest itself as grainy or pixelated disturbances
• Factors that contribute to noise include low light conditions, high ISO settings, and sensor limitations
• Noise refers to a variation of brightness and color that does not correspond to reality

Applying noise with scikit-image

• Import the module and function from skimage.util import random_noise
• Add noise to the image and store in a variable: noisy_image = random_noise(dog_image)
• Show the original and resulting image using the function show_image()
• One type of noise is known as salt and pepper

Reducing Noise

• Denoising is the process of reducing noise
• Types of denoising include total variation (TV), bilateral, wavelet denoising and non-local means denoising

Denoising Techniques in scikit-image

• Total variation filter denoising can be performed withskimage.restoration import denoise_tv_chambolle
• It takes the noisy image and applies total variation filtering denoised_image = denoise_tv_chambolle(noisy_image, weight=0.1, channel_axis=-1)
• Original and denoised versions can be displayed with show_image, showing the noisy and denoised versions
• Another way to reduce noise is by using Bilateral filtering

Using Bilateral Filtering

• Use from skimage.restoration import denoise_bilateral
• Then apply the filter: denoised_image = denoise_bilateral(noisy_image, channel_axis=-1)
• The resulting image from bilateral filter is less smooth than the one from total variation filter, and preserves the edges a lot more

Segmentation and Superpixels

• Segmentation involves partitioning images into regions to simplify and/or change the representation into something more meaningful and easier to analyze.
• Segmentation is used to detect and isolate before a tumor analyzed using computed tomography
• Before recognising a face, it has to be extracted from its background using segmentation
• A single pixel by itself is not a useful representation of an image

Superpixels

• Superpixels are formed by bigger regions or grouped pixels, allowing the exploration of logical meanings in an image
• A superpixel is a group of connected pixels with similar colors or gray levels, and carry more meaning than their simple pixel grid counterparts.

Superpixel Benefits

• Superpixels result in more meaningful regions and computational efficiency

Segmentation Types

• Segementation can be supervised for the kind of thresholding in which a specific value is specified
• Segmentation can also be unsupervised, and requires no prior knowledge, and algorithms automatically subdivide images into meaningful regions (e.g. Otsu)

Unsupervised Segmentation

• SLIC (Simple Linear Iterative Clustering) segments the image using a machine learning algorithm.
• K-means clustering is used to segment the pixel values of the image
• SLIC separates them into a predefined number of sub-regions

SLIC Implementation

• Import the modules skimage.segmentation and skimage.color, importing slic and label2rgb from each library, respectively
• The segments are obtained with segments = slic(image)
• Put the segments on top and compare: segmented_image = label2rgb(segments, image, kind='avg')
• The number of segments can be controlled using: segments = slic(image, n_segments=300)

Finding Contours

Contours

• A contour is a closed shape of points or line segments, representing the boundaries of objects
• Contours can be used to: Measure size, classify shapes, and determine the number of objects
• Binary images, created through thresholding or edge detection, are usefull

Finding Contours with Scikit-image

• Image must be transformed to 2D greyscale with image = color.rgb2gray(image)
• Image must be binarized
• Obtain the thresh value with the threshold_otsu method on the image
• Apply a threshold thresholded_image = image > thresh,
• Import the measure module from skimage, and then use find_contours
• Find contours at a threshold constant value: contours = measure.find_contours(thresholded_image, 0.8)

Constant Level values

• The level value varies between 0 and 1; the closer to 1 the more sensitive the contour detection method will be, resulting to more complex contours will be detected
• Programmers must find the value that best detects the specific desired contours

Identifying contours Steps

• Make the image grayscale: image = color.rgb2gray(image)
• Obtain the optimal thresh value of the image: thresh = threshold_otsu(image)
• Apply thresholding and obtain binary image thresholded_image = image > thresh
• Find contours at a constant value, such as contours = measure.find_contours(thresholded_image, 0.8)

Contour Shapes

• Contours are a list of (n,2) ndarrays

Explore image restoration techniques in Python using scikit-image. Focus on image reconstruction through inpainting to fix damaged images and remove unwanted elements. Use masks to identify damaged regions and restore images using the inpaint.inpaint_biharmonic function.