Practical Image and Video Processing Using MATLAB Oge Marques (read clockwise)

Mind map of Practical Digital Image Processing with MATLAB (Oge Marques)

Get Started. It's Free
or sign up with your email address
Practical Image and Video Processing Using MATLAB Oge Marques (read clockwise) by Mind Map: Practical Image and Video Processing Using MATLAB Oge Marques (read clockwise)

1. Chapter 1 Introduction and Overview

1.1. Image

1.1.1. Visual representation 2D A projection of a 3D real-world item

1.1.2. Digital Finite number of points (pixels) Monochrome Color Alternative representations

1.2. Image processing

1.2.1. Modifying digital images with computers Multidisciplinary Mathematics Physics Computer science Computer engineering Optical engineering Electrical engineering Pattern recognition Machine learning Artificial intelligence Human vision research

1.2.2. Scope Low level Primitive operations Mid level Extract attributes High level Analysis Interpretation

1.2.3. Examples Sharpening: enhance edges and fine details Example Noise removal Example Deblurring Example Edge extraction Example Preprocessing step to separate objects from one another Binarization Example To simplify and speed up interpretation Blurring Example To minimize the importance of textures and fine details Contrast enhancement Example To help humans and other computer tasks, e.g. edge extraction Object segmentation and labeling Example Prerequisite for most objection recognition and classification work

1.3. Components of a digital image processing system

1.3.1. Hardware Acquisition devices Capture Processing equipment Modify and analyze Display device Show Hardcopy device Show Storage device Preserve

1.3.2. Software

1.4. Machine Vision Systems

1.4.1. Overview (without AI/ML)

1.4.2. Acquisition Input: real-world object Output: digital image

1.4.3. Preprocessing Input: digital image Output: corrected/improved digital image

1.4.4. Without AI/ML Segmentation Input: digital image Output: partitioned image Feature extraction Input: preprocessed, segmented image Output: encoded image contents (feature vector) Classification Input: feature vectors Output: human-useful label

1.4.5. With AI/ML (trained model) Classification Input: preprocessed image Output: human-useful label

1.5. Human Vision System (HVS) vs. Machine vision System (MVS)

1.5.1. Storage HVS: large database of images accumulated over a lifetime, mapped to high-level semantics MVS: large storage possible, but not much context (semantics)

1.5.2. Speed HVS: very high speed MVS: fast, but not as fast as HVS (not real time)

1.5.3. Working conditions HVS: can cope with poor lighting conditions, confusing perspectives, etc. MVS: needs good lightning conditions, confused by extraneous objects, angles, etc.

2. Chapter 2 Image Processing Basics

2.1. Digital Image Representation

2.1.1. Basic formats raster (bitmap) 2D matrix of numbers f(x,y) = intensity or gray level at pixel (x,y) Pros/cons The book covers this format vector Drawing commands Pros/cons

2.1.2. Binary (1-bit) 1 bit per pixel 0=black, 1=white (usually) Small size Suitable for simple graphics, line art

2.1.3. Gray-level (monochrome) 8 bits per pixel 256 levels of gray Relatively compact, subjectively good quality

2.1.4. Color 24-bit RGB (red, green, blue) 3 x 2D arrays (one per color) 32-bit RGB + alpha Additional 8 bits for alpha (transparency) for each pixel Used for image editing effects Indexed color For compatibility with older hardware Pointers to a color palette (usually 256 colors)

2.2. Compression

2.2.1. Why Raw image representations are large

2.2.2. Lossy Tolerable degree of deterioration General purpose photographic images

2.2.3. Lossless Full quality Line art, drawings, facsimiles, space images, medical images

2.2.4. Compression rate bpp -> bits per pixel

2.3. File formats

2.3.1. General File header Image height, width, bands, bpp, format signature, ... Pixel data (often compressed)

2.3.2. Examples BIN, PPM, JPEG, GIF, PNG, TIFF

2.4. Terminology

2.4.1. Topology Fundamental image properties Usually done on binary images Number of occurrences of an object Number of separate (not connected) regions Number of holes in an object ...other examples

2.4.2. Neighborhood Pixels surrounding a pixel 4-neighborhood 8-neighborhood diagonal-neighborhood

2.4.3. Adjacency In relation to two pixels (p and q) 4-adjacent: p and q are 4-neighbors or each other 8-adjacent: 8-neighbors of one another mixed- or m-adjacency: eliminate ambiguities in 8-adjacency

2.4.4. Path In relation to two pixels (p and q) 4-path sequence of pixels from p to q where each pixel is 4-adjacent to its predecessor 8-path same, but now each pixel is 8-adjacent to its predecessor Example (expand node to see image) Example image

2.4.5. Connectivity 4-connected there is a 4-path between p and q 8-connected there is an 8-path between p and q

2.4.6. Components Set of pixels connected to each other 4-component 4-connected 8-component 8-connected

2.4.7. Distance In relation to coordinates of two pixels p and q Euclidean sqrt( (x1-x0)^2 + (y1-y0)ˆ2 ) Manhattan (city block) |x1-x0| + |y1-y0| Chessboard max(|x1-x0|, |y1-y0|) Example (expand node to see image) Example picture

2.5. Operations

2.5.1. In the spatial domain Arithmetic/logical operations on the original pixel value Global (point) operations Entire image is treated uniformly: same function applied to all pixels New pixel value is a function of old pixel value Example: contrast adjustment Neighborhood-oriented (local, area) operations Pixel by pixel, typically using a convolution operation New pixel value is a function of its value + neighbors Example: spatial-domain filters (blur, enhance, find edges, remove noise) Combining multiple images Two or more images are combined arithmetically or logically Example: subtract one image from another to detect differences

2.5.2. In a transform domain Transform = convert a set of values to another set of values, creating a new representation for the same information Image undergoes a mathematical transformation (Fourier transform, discrete cosine transformation) From spatial domain (original image) to transform domain (new representation) Examples: frequency-domain filtering

3. Chapter 5 Image Sensing and Acquisition

3.1. Requirements

3.1.1. Illumination (energy) source Electromagnetic raidation

3.1.2. Imaging sensor Converts optical information into electrical equivalent

3.2. Types of images

3.2.1. Reflection images Radiation reflected from the surface of objects

3.2.2. Emission images Objects are self-luminous Visible or invisible raidation

3.2.3. Absorption images Radiation passes through an object Provides information about internal structure (e.g. X-ray)

3.3. Light and color perception

3.3.1. Radiance Physical power Expressed as spectral power distribution (SPD)

3.3.2. Human perception of light Brightness "An area appears to emit more or less light" Luminous intensity Perceptual (cannot be measured) Hue "An areas appears similar to one of the perceived colors (red, green, blue, or combination" Dominant wavelength of the SPD Saturation "The colorfulness of an area judged in proportion to its brightness" A description of the whiteness of the light source The more the SPD is concentrated at one wavelength, the more saturated is the associated color Adding white light (all wavelengths) causes color desaturation Perceptual (cannot be measured) Visualization of the concepts Visualization 1 Visualization 2

3.4. Image acquisition

3.4.1. Sensors Covert electromagnetic energy to electrical signals Types CCD (Charge-coupled devices) CMOS (complementary metal oxide semiconductor) Characteristics Nominal resolution Field of view

3.4.2. Camera optics Characteristics Magnification power Light gathering capacity Aberrations PIncushion distortion Barrel distortion

3.5. Digitization

3.5.1. From analog to digital representation Results in a pixel array Monochrome: intensity Color: color values

3.5.2. Processes Sampling Time or space Usually done before quantization Measure the value of a 2D function (height/width) at discrete intervals Rate: number of samples across height and width Pattern: physical arrangement of the samples Nyquist criterion Illustration Quantization Amplitude Replaces a continuous function with a discrete set of quantization levels Illustration

3.5.3. Resolution Spatial Density of pixels in an image Gray-level Smallest change in intensity level that the HVS can discern

4. Chapter 6 Arithmetic and Logic Operations

4.1. Arithmetic operations

4.1.1. Addition Blend images

4.1.2. Subtraction Detect differences between images

4.1.3. Multiplication and division Brightness adjustment

4.2. Logic operations

4.2.1. Pciture

5. Appendix A Human Vision Perception HVS (human vision system)

6. Chapter 7 Geometric Operations

6.1. What they are

6.1.1. Modify the geometry of an image by repositioning pixels

6.1.2. Modify the spatial relationship between groups of pixels

6.2. Components

6.2.1. Mapping functions Specify new coordinates in the output image for each pixel in the input image Spatial transformation equations: f(x,y) --> g(x',y') Often separate functions for x and y x' = T(x,y), y' = Ty(x,y) Affine transformations Linear combinations of x and y Preserves line parallelism, but not angles and distances Examples Translation Scaling Rotation Shearing

6.2.2. Interpolation methods Compute new values for each pixel Approaches Forward mapping (source to target) Backward mapping (target to source)

6.3. Examples

6.3.1. Zooming

6.3.2. Shrinking

6.3.3. Resizing

6.3.4. Translation

6.3.5. Rotation

6.3.6. Cropping

6.3.7. Flipping

6.3.8. Warping

6.3.9. Non-linear Twirling Rippiling Morphing Seam carving Image registration