Outline Color Image Processing Motivation and Color Fundamentals Standard Color Models (RGB/CMYK/HSI) Demosaicing and Color Filtering Pseudo-color and Full-color Image Processing Color Transformation Tone and Color Corrections Color Image Smoothing, Sharpening, Segmentation, Edge detection, and Denoising Assignments
Motivation Color is a powerful descriptors that simplifies feature extraction and object identification from a scene HVS is sensitive to thousands of color shades or intensities instead of two dozens of shades of gray Basics Color spectrum of visible light has six broad regions, viz., Violet, Blue, Green, Yellow, Orange, and Red (ViBGYOR) Achromatic light is void of color. Chromatic light spans electromagnetic spectrum from 400 nm to 700 nm. Radiance is the amount of light energy radiates from light source in Watts. Luminance is the energy perceives by viewer in Lumen. Infrared spectrum may have significant radiance but zero luminance Brightness is a subjective measure, depends on the reflectance or absorption characteristics of the observed body
Color Spectrum Color spectrum of white light passing through a prism. Experimented by Sir Isaac Newton in 1666
Absorption Characteristics 65% cones are sensitive to Red 33% cones are for Green 2% cones are for Blue, but very sensitive! Absorption characteristics experimented in 1965.
Primary and Secondary Colors In 1931, CIE (Commission Internationale de l Ecairage International Commission on Illumination) designated three primary color : Blue = 435.8 nm, Green = 546.1 nm, and Red = 700 nm long before absorption characteristics are obtained. Additive mixing of primary color lights provides secondary color of lights: Magenta= Red+Blue, Cyan= Green+Blue, Yellow=Red+Green. Additive mixing primary color lights is used as principle of modern display devices. CRT (cathode ray tube) uses electron sensitive phosphor. LCD (liquid crystal displays) uses thin film transistors (TFTs) to block or pass polarized light. In plasma units pixels are tiny gas cells coated with phosphor.
Primary and Secondary Colors Pigments act like subtractive colors, and hence mixtures of pigments provide common primary color of light. For example, Magenta and Yellow provides Red color.
Characteristics of Color Brightness Achromatic notion of intensity. Hue Dominant wavelength in the mixture of light waves, i.e., dominant color perceived by observer, say, Red Saturation Relative purity or the amount of white light mixed with a hue. For example, pink (red+white) and lavendar (violet+white) are saturated. Degree of saturation is inversely proportional to the amount of white light. Chromaticity Hue and saturation together is the chromaticity. Any color is specified a combination of tristimulus values denotes as X (Red), Y (Green), and Z (Blue)
Characteristics of Color A color is thus specified by trichromatic coefficients x = X X + Y + Z ; y = X x + y Y + Y + Z + z = 1 ; z = X z + Y CIE chromaticity diagram For a given value of x (red) and y (green), a corresponding value of z=1-(x+y), i.e., blue is obtained A color is thus specified by trichromatic coefficients Any triangle inside chromaticity diagram shows all possible colors which may be obtained mixing only three distinct wavelength primary electromagnetic waves. + Z
Color Image Processing Chromaticity Diagram
Color Gamuts CRT Display Printing
RGB Color Model Color space in terms of three primary colors of light, viz. Red, Green, and Blue Hardware oriented models, e.g., monitors, video camera, internet etc. Number of bits used to represent a pixel is called pixel-depth. For a 8-bit image one needs 24 bits in color space 24-bit color cube Coordinate
RGB Color Model Acquiring the RGB image in the reverse process shown Three hidden planes in the cube
Safe RGB Colors In practice, only 216 colors known as allsystem safe colors are used in the internet and represented by two hexadecimal numbers for each color safe color cube
CMY/CMYK Color Model Color space in terms of the primary colors of pigment, viz. Cyan, Magenta, Yellow, or Black Printer or copier those use pigments on paper considers this model. Four-color printing refers to CMYK model Conversion of RGB and CMY model: C M Y = 1 1 1 R G B
HSI Color Model Color space in terms of its Hue, Saturation and Brightness The model is used for interpretation of human understanding of colors This model is used for image processing. The model decouples the intensity component from the color descriptors It is known that Hue purity of color It is also known that Saturation degree of pure color (diluted with white color)
HSI and RGB Color Model Intensity of HSI space is the projection of a point on the vertical line (connected line between black and white, called intensity axis) shown in the RGB space Locus of the color points lie on the plane perpendicular to the intensity axis. Conceptual model in HSI space
Hue/Saturation in HSI Model The color vector perpendicular to the intensity axis creates a triangular or hexagonal shape with the boundaries of the cube or circular shape inside the cube. The saturation is the length of the color vector. Hue is the angle of the vector.
Hue/Saturation in HSI Model Hue and saturation in the triangular or circular color planes
RGB to HSI Model θ if B G Hue H = 0 360 if B > G 1 1 2 [( R G ) + ( R B )] with θ = cos 2 ( R G ) + ( R B )( G B ) [ ] 1 2 Saturation S = 1 3 [ min { R, G, B }] ( R + G + B ) Intensity ( R + G B ) 1 I = + 3
HSI to RGB Model RG Sector ( 0 0 0 H < 120 ) S cos H B = I ( 1 S ); R = I 1 + ( ) ; G = I- + H 3 0 cos 60 ( R B ) GB Sector ( 0 0 120 H < 240 ) H = H 120 0 S cos H R = I ( 1 S ); G = I 1 + ( ) ; B = I- + H 3 0 cos 60 BR Sector ( 0 0 240 H 360 ) H = H 240 ( R G ) S cos H G = I ( 1 S ); B = I 1 + ( ) ; R = I- + H 3 0 cos 60 0 ( R B )
Components of HSI Model (a) (b) (c) (d) (a) Original RGB image. The component images are (b) Hue (c) Saturation (d) Intensity
Image Demosaicing Demosaicing is a process of obtaining a full color image from incomplete color samples A single image sensor associated with the color filter array (CFA) performs the demosaicing A digital camera may store the raw data so that user-defined software for CFA may be chosen instead of built-in firmware. Problems of image demosaicing includes chromatic aliases, zippering (abrupt change in intensity), and purple fringing. Most commonly used CFA for demosaicing is the Bayer Filter.
Bayer Filter Bayer Filter has alternating Red (R) and Green (G) filters in odd rows and alternating Blue (B) and Green (G) filters in the even rows. Green filters are twice since HVS is widely sensitive to Green color! Optical anti-aliasing filter is used between sensor and lens. CFA on pixel array of image sensor Image due to CFA Original Reconstructed (Adobe)
Pseudocolor Image Processing In intensity slicing, a set of intensity levels are coded with a particular color Functional diagram Geometric interpretation
Intensity Slicing Monochrome is color coded Two colors Eight colors
Intensity Slicing Monochrome is color coded for SAR images
Pseudo-color Processing Pseudo-color enhancement by gray level to color transformation Explosive detection
Pseudo-color Processing Two gray level transformation functions for detection of explosives in a luggage in a typical airport Functional diagram
Pseudo-color Processing Pseudo-color enhancement of multispectral images (a)-(d) Images of four bands. (e) First three are treated as RGB components (f) Infrared image is shown in red (a) (b) (c) (d) (e) (f)
Color Image Processing Pseudo-color Processing Pseudo-color enhancement in terms of material deposition, e.g., sulfur content shown in yellow for Jupiter moon Io
Full-color Processing Full-color processing uses independent mask processing of individual RGB components
Full-color Processing Components for full color processing in different color spaces
Color Transformation ( ) General transformation s = T r r,, r For RGB or CMY or HSI n = 3 For CMYK n = 4 i i, L 1 2 n Color mapping s = kr 0 < k < 1 s = r s = r ( ) HSI 3 3 1 1 2 2 ( ) RGB s = kr 0 < k < 1 i = 1,2, 3 i i CMY s = kr + ( 1 k ) ( 0 < k < 1) i = 1,2, 3 i i
Color Transformation Intensity reduction by 33% in three different color spaces using full-color processing
Color Complements Hues directly opposite to color circle are called color complements Complements are used to generate color film negatives
Color Complements (a) (b) (c) (d) (a) Original image (b) Transformation functions for generating complements (c) Resultant image for RGB space (d) Resultant image for HSI space
Color Slicing Selection of a color using a hypercube in a color space Red color in the RGB space using two different set of cubes
Tone and Color Corrections Digital darkroom Allows tone adjustment and color corrections digitally by avoiding traditional wet processing Most common use are photo enhancement and color reproduction in the printing media and compression For adjustment or corrections device-independent color model should be used Many color management system (CMS) consider the CIE L* a * b * or simply CIELAB model CIELAB is proposed in 1976
Color Image Processing Color Image Processing CIELAB Model The color components are * * * b a L = = = W W W W W Z Z h Y Y h b* Y Y h X X h a Y Y h L 200 500 * 16 116 * ( ) + > = 0.008856 116 16 7.787 0.008856 3 q q q q q h where
CIELAB Model X, Y, Z The W W W are the reference white tristimulus values of a perfectly reflecting diffuser under CIE standard D65 illumination, which is defined as x=0.3127 and y=0.3290 in the CIE chromaticity diagram Features of color space Colormetric Colors perceived as matching are encoded identically. Uniform perceptuality Color differences among various hues are perceived uniformly Device independent and gamut encompasses the entire visible spectrum Transformation to another color space is common L* a* b*
CIELAB Model Other features of color space Transformation to another color space is necessary Excellent decoupler of intensity and color for Red-Green and for Green-Blue Tonal adjustment and color corrections are done independently and interactively (in other words sequentially) Three common tonal ranges (or key types) are (i) Highkey (colors at high intensities) (ii) Low-key (colors at low intensities) (iii) Middle-key (lie in between high and lowkeys) Intensites should be uniform in colors or shadows L* a* b* b* L * a *
Tonal Transformation Tonal or contrast correction and corresponding transformation function in RGB color space Middle- Key High- Key Low- Key
Color Transformation Color balancing correction and corresponding transformation function in CMYK color space
Histogram Processing Histogram equalization followed by saturation adjustment in HSI color space
Processing of Color Image RGB Components HSI Components
Smoothing of Color Image (a) (b) (c) Image smoothing by 5 5 averaging mask using (a) RGB components and (b) Intensity component of HSI space. (c) Absolute difference between (a) and (b)
Sharpening of Color Image (a) (b) (c) Image sharpening by Laplacian using (a) RGB components and (b) Intensity component of HSI space. (c) Absolute difference between (a) and (b)
Segmentation of Color Image Approaches for enclosing data regions for RGB vector segmentation
Segmentation of Color Image (a) Original RGB image (b) Segmented region of red color from the image (a) (b)
Color Image Processing Color Image Processing Gradient of Color Image Gradient of color image y B x B y G x G y R x R g y B y G y R g x B x G x R g xy yy xx + + = + + = + + = 2 2 2 2 2 2 Maximum rate of change would be at ( ) = yy xx xy g g g y x 2 tan 2 1, 1 θ The maximum gradient would be ( ) ( ) ( ) ( ) ( ) { } 2 1, sin 2 2, cos 2 2 1, + + + = y x g y x g g g g y x F xy yy xx yy xx θ θ θ
Gradient of Color Image (a) (b) (c) (d) (a) Original color image (b) Color gradient (c) Added result of individual gradients of RGB components (d) Absolute difference between (b) and (c) Individual gradients of RGB components
Noisy Color Image (a) (b) (c) (d) (a) Red (b) Green (c) Blue components of RGB color space for AWGN (d) Noisy color image
Noisy Color Image (a) (b) (c) (a) Hue (b) Saturation (c) Intensity components of HSI color image for AWGN
Color Image Processing Noisy Color Image (a) Color image corrupted by Saltand-Pepper noise (b) Hue (c) Saturation (d) Intensity components of HSI color space (a) (b) (c) (d)
Assignments Problem #1 Problem #2
Assignments Problem #3 Derive the CMY intensity mapping function from its RGB counterpart. Problem #4