Image Processing Basics Quiz

Computer Vision Lecture Notes

Computer vision is an introductory course to image processing and computer vision.
The main objectives of the course are to identify the difference between image processing and computer vision, identify different data structures for image analysis and identify and understand image features.
Computer vision is a subfield of AI focused on getting machines to see like humans.
The goal of computer vision is bridging the gap between pixels and meaning.
Humans are very good at interpreting visual input.
Visual perception is difficult for machines
To "see", humans use sensing and interpreting visual data (images or videos) to understand what is where, what actions are being performed and future predications.
Visual perception is complex.
Optical illusions show that even humans find it hard to correctly interpret images.
Computer vision is about getting computers to analyze and understand images and videos.

Image Processing

Image processing is to either improve a picture for human interpretation or make it easier for machines to understand.
Humans prefer images that are sharp, clear and detailed.
Machines prefer simple and uncluttered pictures.
Examples of image processing include enhancing edges, removing noise, and removing motion blur
Image processing is one part of computer vision.
Computer vision uses image processing algorithms.
An image represents something (person, scene, or object)
An image consists of a set of points called pixels.
Each pixel has a particular brightness.
Examples of image types are binary images, grayscale images, and color images.
Binary images are black or white pixels.
Grayscale images have shades of gray.
Color images consist of amounts of red, green and blue in each pixel.
There are 16,777,216 different possible colors in an image.
Pixels have 24 bit colour values.
A digital image is a matrix of intensity values.
Basic transformations include shifting pixel values, and combining pixel values from surrounding pixels.
Image transformations are categorized into three classes, based on the information needed to complete a transformation. These are point operations, neighbourhood processing, and transformations.
Point operations modify a pixel intensity without considering surrounding pixels.
Neighbourhood processing modifies a pixel by considering the values around it.
Transforms modify every pixel simultaneously in a large block.
Arithmetic operations: Examples include adding (lighten) and subtracting (darken) constants to pixels, and multiplying pixel values by constants, to modify images. Negating an image (i.e. finding the complement) is also a possible arithmetic operation

Other Computer Vision Applications

Image Retrieval: locating images without words in large databases.
Optical Character Recognition (OCR): turning scanned documents into text.
License Plate Recognition: processing images to extract characters.
Face Detection: detecting faces in images or videos by using digital cameras.
Smile Detection: instantly detecting smiles (used in digital cameras).
3D Model Building: creating 3D models using images, such as reconstructions of famous buildings.
Forensics: using images in legal contexts (in the detection of crime or for legal purposes).
Biometrics: using images for identifying people based on biometrics like fingerprints or facial recognition and identification.
Vision Based Interaction and Games: the use of camera technology in games and other forms of interaction.
Smart Cars: using cameras to drive cars autonomously.
Medical Imaging: using technology (like MRI, CT) to create 3D images for medical purposes.
Computer Vision is difficult because of a number of reasons. These include viewpoint variations, illumination, intra-class variations, motion, scaling, and background clutter

Image Processing-Lecture(2)

Types of Images: binary, grayscale, colour.
Switching Between Formats: - RGB to Grayscale: average the red, green and blue components of each pixel. - Grayscale to binary: apply a threshold of a certain value.

Geometric transformations

Rotation: a transformation to rotate an image.
Resizing or scaling: a transformation used to change the size of an image.
The two-step process involves spatial transformation (relocating known pixel values), where the new location of each pixel is calculated after scaling, and interpolation to determine the transformed image's values at integer pixel locations. Examples include Nearest neighbour interpolation, and Bilinear interpolation

Neighbourhood Processing

Image Filtering: a transformation that combines several pixels in an image based on a function to modify an image.
Filters are used to modify images using some function of a certain neighbourhood of each pixel in the image.
Linear filters will use an algebraic approach (a weighted sum) with a kernel over all pixels.
In the spatial domain, linear filtering is implemented by finding the convolution of a filter kernel over an image.
A convolution involves flipping rows and columns of a kernel. Multiplying pixels in the range of a kernel by the corresponding element of the flipped kernel, summing all products will determine the new pixel value

2D Convolution

Used to convolve image I with filter kernel K.
Involves flipping rows and columns of the kernel, multiplying each kernel pixel within the range of the kernel with the corresponding flipped kernel element, summing these products to calculate a central pixel.
Performed over every pixel in the image.

Histograms

Given a gray scale image, its histogram consists of the histogram of its gray levels. A graph is used to visualise the histogram. It conveys information about the appearance of the image.
Examples include a dark image (concentrated at the low-end of the histogram), a uniformly bright image (concentrated at the high-end of the histogram) and a properly contrasted image (spread out across a large portion of the histogram).
Histogram stretching (contrast stretching) and histogram equalization are used to enhance contrast in an image by spreading out the image histogram (using a function).

Canny Edge Detectors

Used to extract meaningful structural information.
Probably the most widely used edge detector in computer vision.
Steps include: - Smoothing the image using a Gaussian filter. - Calculating the gradient magnitude and orientation. - Performing non-maximum suppression: thinning multi-pixel wide ridges. - Applying hysteresis thresholding.

Feature Extraction

Corners: areas where two different strong edge directions occur in an image. - Using Harris corner detection as an example, the process involves calculating "Error" for each small shift window on each pixel of the image, summing up the squared differences between the pixels (using sum of squared differences (SSD)). Calculating values for A, B, C, helps classify the points as edges, flat, or corners by considering eigenvalues of matrix M.