Understanding Image Files and Formats

• Images are visual representations of real-life objects in a 2D form.
• Images consist of a grid of pixels, each with a numerical value representing color or intensity.

Image Files

• An image file is an encrypted file containing image data and metadata.
• EXIF metadata is automatically generated by cameras and smartphones, including camera details, image settings, date & time, flash information, GPS coordinates, image orientation, resolution, and DPI.
• IPTC metadata is used for professional photography or journalism, including title, description, copyright information, location data, and photographer information.
• XMP metadata is frequently used in Adobe software for custom metadata, software used, editing history, color profile, and custom tags.

Image File Formats

• JPEG: Joint Photographic Experts Group, uses lossy compression for photographs and paintings
• TIFF: Tagged Image File Format, lossless compression for document storage
• GIF: Graphics Interchange Format, bitmap image format supporting animation
• BMP: Bitmap, independent of display and lacks compression, for use in Windows
• PNG: Portable Network Graphics, lossless data compression supporting different color spaces, used for image transfer over the internet
• WebP: Format developed by Google, uses lossless and lossy compression, optimized for small image size comparable quality to JPEG, and used for stickers in messaging apps
• SVG: Scalable Vector Graphics, used for webpage development, behaviors and images defined in XML format, which can be searched, indexed, and compressed.

Elements of an Image File

• Pixels
• Dimensions
• Resolution

Pixel

• A pixel is the basic unit of an image, represented by numerical values depending on the color model used.
• Grayscale images use a single intensity value ranging from 0 (black) to 255 (white).
• RGB images use three values (Red, Green, Blue) for each pixel, each ranging from 0 to 255.

Dimension

• An image dimension refers to the number of pixels in the first and second dimensions.
• An image dimensions represented as 500x400, signifies 500 pixels by 400 pixels, totaling 200,000 pixels.

Resolution

• Image resolution is the number of pixels per inch (ppi) in an image.
• Higher resolution images have more detailed color distribution, resulting in increased file size.

Image Resolution: Units of Measurement

• Pixels per inch (PPI) measures pixel density within a digital image.
• Dots per inch (DPI) measures ink droplets or toner dots a printer lays down per inch of paper.

Image Compression

• Image compression reduces file size while maintaining acceptable quality.
• Types of image compression include lossless and lossy compression.

Lossless Compression

• Lossless compression reduces file size without compromising image quality.
• During decompression, all information is restored.

Lossy Compression

• Lossy compression may result in some data loss.
• This compression prioritizes saving space over maintaining accuracy of the retrieved file.

Lossy Compression vs Lossless Compression

• Lossy Compression: smaller file size with significant reduction, quality loss with visible artifacts, best for web images, photos, social media, email, examples include JPEG, WebP, HEIF/HEIC
• Lossless Compression: larger file size with less reduction, no loss in quality, best for archiving and professional photography, examples include PNG, TIFF, GIF, and BMP

Color Spaces

• Color space refers to the organization of colors in an image in a specific format.
• Color model refers to the way in which a color is represented.

Color Model Examples

• RGB: Red, Green, Blue
• CMYK: Cyan, Magenta, Yellow, Black
• XYZ: color in the x, y, and z dimensions
• HSV/HSL: hue, saturation, and value/hue, saturation, and lightness
• LAB: luminance, and green-red and blue-yellow color components
• LCH: lightness, chroma, and hue
• YPbPr: green, blue, and red cables
• YUV: brightness and chroma, or color
• YIQ: luminance, in-phase parameter, and quadrature

Color Models and Light

• Light is an electromagnetic wave and differentiated by the naked eye.
• Human eyes detect light from 700 nm to 400 nm, with infrared and ultraviolet light not being visible.

Color Models: Purpose and Orientation

• Color models facilitate the specification of colors in a standard way.
• They are oriented toward hardware or applications.
• Hardware-oriented models: Color monitor (RGB), Color printer (CMYK), Color TV broadcast (YIQ)
• Image manipulation: HSI, HSV
• Image processing: RGB, YIQ, HSI

Color Models: Processes

• Additive processes create color by adding light to a dark background (monitors).
• Subtractive processes use pigments or dyes to selectively block white light (printers).

Color Models: Attributes

• Intensity is the strength or purity of a color.
• Brightness is the lightness or darkness of a color.
• Luminance (Y) radiant power weighted by a spectral sensitivity function that is characteristic of human vision.
• Lightness (L*) is the nonlinear perceptual response to luminance, roughly logarithmic.

RGB Color Model

• RGB is the most used color model in digital image processing.
• It can be converted vice versa with the XYZ color model or to HSI/HSL.
• In the red, green and blue system the color solid generated is a bounded subset of the space generated by each primary.
• Using an appropriate scale,the space can be normalized, so that the maximum is 1.
• The RGB color solid is represented as a cube, with black defined as (0,0,0) and the system's brightest white at (1,1,1).
• The color subspace of interest is the cube, colors are defined by vectors extending from the cube origin and normalized in the range of 0-1.

XYZ Color Model

• The XYZ model is a device-independent model useful for consistent color representation across different devices.
• Critical for color management purposes.
• It has a significant role in image processing.

RGB to XYZ Conversion

• RGB color space can be transformed into XYZ color space using matrix multiplication where RGB and XYZ values are in the range [0, 1].
• The formulas for this are:
  • R = 3.2405𝑋 -1.5372𝑌 -0.4985𝑍
  • G = -0.9693𝑋 1.8760𝑌 0.0416𝑍
  • B = 0.0556𝑋 -0.2040𝑌 1.0573𝑍
  • X = 3.2405R -0.9693G 0.0556B
  • Y = -1.5372R 1.8760G -0.2040B
  • Z = -0.4985R 0.0416G 1.0573B

HSI Color Model

• The HSI (hue, saturation, intensity) model decouples intensity from color-carrying information in an image.
• The HSI model is an ideal for developing image processing algorithms based on color descriptions that are natural and intuitive to humans.
• Hue: A color attribute that describes a pure color, represented as an angle on the color wheel (0° to 360°).
  • 0° = Red, 60° = Yellow, 120° = Green, 180° = Cyan, 240° = Blue, 300° = Magenta
• Saturation: Measure of how much a pure color is diluted with white light, measured as a percentage, with 0% meaning no saturation (gray) and 100% meaning full saturation (pure color).
• Intensity: Refers to the brightness of the color, often represented as a percentage, where 0% is black and 100% is the brightest version of the color. Also referred to as "Value."

RGB to HSI Conversion

• Formulas for conversion:
  • 𝐻 = cos^−1 [((𝑅′ − 𝐺′) + (𝑅′ − 𝐵′)) / ((𝑅′ − 𝐺′)^2 + (𝑅′ − 𝐵′)(𝐺′ − 𝐵′)) ^1/2]
  • 𝑆 = 1 − [3 / (𝑅′ + 𝐺′ + 𝐵′)] [min(𝑅′, 𝐺′, 𝐵′)]
  • 𝐼 = 1/3(𝑅′ + 𝐺′ + 𝐵′)
  • where 𝐻 = 360° − 𝐻, if (𝐵′/𝐼) > (𝐺′/𝐼).
• Hue (H) isn't defined when saturation (S) is zero and Saturation (S) in undefined if intensity (I) is zero.

Color Model Comparison

• RGB: Components - Red, Green, Blue; Characteristics - Additive model, intuitive for screens, directly represents pixel colors; Use Cases - Image display, basic editing, machine learning; Limitations - Not perceptually uniform, sensitive to lighting variations.
• HSV: Components - Hue, Saturation, Value; Characteristics - Separates color from brightness; Use Cases - Color segmentation, artistic adjustments; Limitations - Not perceptually uniform, small changes in one component may not appear consistent.
• HSL: Components - Hue, Saturation, Lightness; Characteristics - Similar to HSV but uses lightness; Use Cases - Color grading, graphics design; Limitations - Non-uniform, computationally similar to HSV.
• YCbCr / YUV: Components - Luminance (Y), Chrominance (Cb, Cr); Characteristics - Separates brightness from color information; Use Cases - Compression, skin detection, video processing; Limitations - Compression, skin detection, video processing.
• Lab: Components - Lightness (L), a (Green-Red), b (Blue-Yellow); Characteristics - Perceptually uniform, approximates human vision; Use Cases - Color correction, perceptual color comparison; Limitations - Computationally complex, less intuitive for basic image processing.
• Grayscale: Components - Intensity; Characteristics - Single-channel model; Use Cases - Edge detection, thresholding, preprocessing; Limitations - Loses all color information.
• CMYK: Components - Cyan, Magenta, Yellow, Black; Characteristics - Subtractive model; Use Cases - Print media; Limitations - Not suitable for screens.

Gamma Correction

• Gamma defines the relationship between a pixel's numerical value and luminance.
• Gamma correction adjusts the image's brightness levels to match perception, compressing darker tones and expanding lighter ones.
• Also referred to as gamma encoding or gamma compression.
• Without Gamma Correction: dark areas are crushed with lack detail and appear overly dark, bright areas are blown out and washed out with lack of subtle gradations, low or flat overall contrast, and doesn't match human perception.
• With Gamma Correction: dark areas have improved and more visible details, bright areas retain highlights and show more nuance, increased overall contrast, and better matches human perception.

Image Histograms

• Image histograms show the distribution of discrete intensity levels in the range [0, L - 1], using a discrete function assigning each intensity level to its number of pixels.
• A histogram has two axes: x and y.
  • X-Axis (Pixel Intensity Values): Represents pixel values, typically ranging from 0 (black) to 255 (white) in an 8-bit grayscale image.
  • Y-Axis (Frequency of Pixels): Number of pixels for each intensity value.
• In a color image, separate histograms exist for the Red, Green, and Blue (RGB) channels.
• The x axis of a histogram represents of the frequency of each intensity, while the x axis represents the range of pixel values.

Applications of Histograms

• Image enhancement (brightness and contrast adjustment, histogram equalization)
• Thresholding
• Object detection
• Feature extraction

Histogram Processing Techniques

• Histogram Sliding
• Histogram Stretching
• Histogram Equalization

Histogram Sliding

• The complete histogram is shifted towards rightwards or leftwards.
• Each pixel intensity in the image is adjusted by adding or subtracting a constant value.
• Increasing brightness in histogram sliding: slide its histogram towards whiter portion
• Decreasing brightness histogram sliding: subtracting constant 80 to each pixel intensity

Histogram Stretching

• Histogram stretching (contrast stretching) enhances an image's contrast by expanding the range of pixel intensity values.
• Contrast is the difference between maximum and minimum pixel intensity.
• Given the matrix of the image as:
  • 100 100 100 100 100
  • 100 100 100 100 100
  • 100 100 100 100 100
  • 100 100 100 100 100
  • 100 100 100 100 100
    • Then:
      • Maximum value in this matrix is 100.
      • Minimum value in this matrix is 100.
      • Contrast = 100 – 100 = 0

Histogram Equalization

• Histogram equalization is used for equalizing all the pixel values of an image and done by transforming uniform flattened histogram.
• Dynamic range of pixel values are increased for an equal count of pixels at each level which produces a flat histogram with high contrast image.
• While stretching histogram, the shape of histogram remains the same whereas shape changes for Histogram equalization.

Histogram Technique Summary

• Histogram Sliding: Adjusts brightness up or down, no change to the shape
• Histogram Stretching: Expands contrast by adjusting intensity range, linear shape
• Histogram Equalization: Redistributes intensity for uniform contrast, non-linear shape