apmiwv>rin COLOR IMAGE PROCESSING 5.1_ INTRODUCTION Till now we have discussed the gray level images where a pixel value at any image point can be represented by a single-valued function with a range of 0 to 255 in a 8-bit system. Image may be multivalued also by making more than measurement at a point. These types of images can be grouped into two major types: 1. Multi-spectral : As the name suggests, measurement corresponds to different spectral bands of frequencies (‘or’ wavelength) of electromagnetic wave. For example, color imaging. 2. Multi Model : The measurement of different physical properties of any single object ‘or’ its parts are concept of multi-model imaging. Different sensors are employed to extract multi- model characteristics. For example, X-ray, CT-Scan, MRI, ete. Now here we will study multi-spectral-color imaging in detail. 5.2_ ADVANTAGES OF COLOR IMAGE PROCESSING () By color image processing we can identify object easily. (i) Human can discern thousands of colors shades and intensities, compared to only two dozen shades of gray. 5:3 CATEGORIES OF COLOR IMAGE PROCESSING (@) Full color processing : In this method, we acquire color image directly by TV Cameras, ‘or’ colorscanner. Cea. sl (6) Pseudo-color processing : In this category, we assign a color to a particular monochrome intensity ‘or’ range of intensities. $4 COLOR FUNDAMENTALS Three basic quantities are used to describe the quality of a chromatic light source: 6364 Digital Image Processing (@) Radiance : Radiance is the total amount of energy that flows from light source. Itis measur Due to this sensitivity of human eyes with respect to colors. Red, green and blue are the primary colors. in watts. (©) Luminance: Luminance gives the measure of energy that a observer perceives from the sour, Itis measured in lumens. (0 Brightness : Itis a subjective descriptor that is practically impossible to measure. 5.5 PRIMARY COLORS Experimentally it has been seen that 6-7 million cones in human eyes can be divided into three basic sensing categories, corresoponding roughly to red, green and blue. 65% cons — sensitive for red. 33% cons — sensitive for green and 2% cons — sensitive for blue. | Absorption Fig. 5.1. Primary color spectrum. 5.6 SECONDARY COLOR Primary colors can be added to get secondary colors of light. Magenta — Red + Blue Cyan— Green + Blue Yellow — Red + Green Magenta Red YZ, Mp2 White cyan q Yellow Fig. 5.2. Primary and secondary color.Color Image Processing 65 7 PRIMARY AND SECONDARY COLORS FOR PIGMENTS 5. Justopposte to thesimple technique. lust oppos _s Primary color— Magenta, cyan, yellow “5 secondary color—Red, blue and green, 58 CHARACTERISTICS THAT ARE USED FOR DIFFERENTIATING DIFFERENT 3. COLORS (@) Brightness : It is notation for intensity. It cannot be practically measured. () Hue: Dominant wavelength in a mixture of light waves (©) Saturation :It gives the measurement of degree to which pure light is diluted by white light. 1 White light added Pure color has high saturation and pink color means mixed color has less saturation. Saturation « Hue + saturation = called chromaticity. 5.9 COLOR MODELS The purpose of color model is to facilitate the specification of colors in some standard, generally accepted way. Thusin color model, we use coordinate system so that each color can be represented by a point. These are different types of color models we can choose any one according to our requirement. (®) RGB Model : Used for color model and broad class of color models corners. () CMY ‘or’ CMYK Model : Used for color printing (9 HSI Model : Used for human description and interpretation of colors. It decouples the colors and gray scales in an image. iron (Site $9.1 AY 3 The RGB-Color Model Inthis model, we use cartesian coordinate system for presentation of color. | RGB are at three points at comers and atbasic scale. ) Magenta, yel © Blackis situ White shows ANlother col ftom origin, low and cyan are at three remaining points at corner. ‘ated at origin and White is situated at the farest point. The line joining black and 'S gray scale, lors are inside the cube in the form of points and are defined by vectors extending (0 Pixel depth: Number Of pixels used to represent a pixel is called pixel depth. The tal application, we require 8 bits for presenting.a color. 's, for RGB, weneeds = 8 x3 = 24 Bits So total number of colors = 224 = 16, 777, 216.66 Digital Image Processing Magenta }—> Gray scale Black R Yellow Fig. 5.3 The RGB model in co-ordinate system. (P) But this large number of colors are practically not usable that why we use only 256 colors. “The set of colors that can be used in stead of whole colors is called subset of colors. Thatés also calle safe RGB colors”. 5.9.2. The CMY and CMYK Color Model Alternatively colors can be represented by secondary colors (‘or’ primary colors of pigment). By RGB, we can get CMY by 1] [R CMY = }1/-|G i B} This is generally used for color printing, But the output color black is muddy black that’s why W® add black in this model, so called CMYK color model. 5.9.3 HSI Color Model “HSI model decouples the color and gray levels. It is the model that will provide information about image tit is easily described and interpreted”. Black Fig. 5.4 HSI color model. mcecy eae abe od [es pane ett black and white points called intensity 2x! e nsity of any color ; we wil i intens! axis and containing that color point. : Regeaes ce ConenclsColor Image Processing 67 = In this plane, the saturation will increase, if the distance f ity axi ; from th ea ie intensity axis has been — This plane will contain same hue at every plane at verticle axis. Point. To get different hue, we have to rotate the Now tocalculate the hue at any point just se the different representation of color model as given below Green Yellow ys white eg cyan oe Red yan Red Green Yellow Blue Magenta Blue Magenta Fig. 5.5 Hue color model and calculation of Hand S. By this figure, it should be clear that angle between primary colors is 120° and also the angle between secondary colors is 120°. Now we have taken red color as reference and hue increases counter clock wise from there. Saturation is calculated by the distance from reference point. 5.10 CONVERSIONS BETWEEN COLOR MODELS 5.10.1 Converting Colors from RGB to HSI Given an image in RGB color formate, the H components of each RGB pixel is obtained by equations: 6 if BsG - (Gal) He (ee if eee 1 +[(R-G)+(R-B)] With @= cos) —2 | 7 aoe. [Ro +(R-B)(G-B)] The saturation components given by 3 i 63) =1-—3_— [min ®,G,B)] S=1-Rrese | Finally, the intensity components given by 64 1-1 (R+G+8) Ca o en meas- It is assumed that RGB values have been normalized to range [0, 1] and angle @ has bet ured w.r.t. red axis of HSI space.= 68 Digital Image Processing 5.10.2 Converting colors from HSI to RGB RG sector (0° < H < 120°): When His in this sector, the RGB components are given by equation B=1(1-S) Rett ScosH ~ [cos (60° H) (3) 1-(R+B). GB Sector (120° < H < 240°) : For this sector, H=H-120° Now, R=1(1-S) S.cos H Sea |r ooe : iy fe otal 69 B=1-(R+G) BR Sector (240° < H< 360°): H=H - 240° Now, G=1(1-S) S.cos H a BL aoe (5.7) [ ses 6. R=1-(G+B) 5.11__PSEUDO COLOR IMAGE PROCESSING “In this process, we simply assign a color to a gray level that’s why called false coloring ‘or’ Pseudo-colot image processing” 5.11.1 Intensity Slicing One of the most popular technique is intensity slicing and then color coding. Now take image a 3-D function. Now place a parallel plain to the coordinate plane of image £ (xy) > (aray - level axis) Fig. 5.6 Intensity slicing.Color Image Processing 69 Now we may assign two different colors; if the gray levels are above then plane, we will assign one olor. If gray level below then plane, we will assign second color. The gray level atthe plane ‘teelf, that can be assigned arbitrary any color out of these two colors [fwe want to take more than one color, we can move parallel plane up and down parallely. Wecan represent the slicing technique by the following figure: Colors Fig. 5.7 _ Representation of intensity slicing. 5.11.2 Grey Level to Color Transformation One possible and easy method of color transformation is that gray levels of an image are fed sepa- rately to red, green and blue transformations and then fed these three colors to different channels of color television monitor. Now by these channels of color television, we used to generate composite (aad Red fy (xa a ee |e oa en fer en image. Blue =» fa (xy) transformation Fig. 5.8 Color transformation of an image. “Note that this transformation is a function of gray levels only, not the positions of gray levels". 3 . , or 1g lot slicing is a special case ofthis transformation but this technique provides smooth, n ‘near flexible functions of transformation. ie Th i F e i ' i i e time itis req technique discussed just till now is applied to single image, but some MN stor. : , eee anumber ofimages ina single composite signal. Then we follow the ion :70 Digital Image Processing fen > a) (x.y) [> he (xy) fy) Additional ee Processing ax{osy) fy) [>> hey) Fig. 5.9 Color transformation of images. 5.12 COLOR TRANSFORMATION “In this transformation, we will discuss the process of all the components of color image within single colo model”. (till now we have discussed in the form of RGB, HSI and CMYK). 5.12.1 Formulation We know for transformation of an image suy=TKeY) Now f(x,y) = color image and g(t, y) = output processed color image. and T= transformation operator onf. -» (68) Now (2,y) isnota single value but triplet ‘or’ quartets value for color image. We again know that p= Ty ye Pay eeey My) FED, 2y ere 64) Here n denotes the number of components to present a image. For RGB, n= 1,2,3,as RGB for CMYK, n= 1,2,3,4.as CMYK G10) i= denotes the transformation function to produce aS, Ingeneral if we wantto (@ modify the intensity ofimage g(uW=kf(x,y) O3)o ‘ 5;= | h 5 otherwise where w= width of window (cube) of RGB. This natural color is middle gray ‘r’ (0.5, 0.5, 0.5). (6.15) 5.13 COLOR IMAGE SMOOTHING AND SHARPENING ‘As we know that full-color images have at least three components so color pixels are vectors. For example, in the RGB system, each color point can be interpreted as a vector extending from origin to that point in RGB co-ordinate system. Let ‘C’ represents an arbitrary vector in RGB color space. Ce] [R c=|C,|=|6 6:18) cy] [B This equation indicates that the components of Care simply the RGB components of color ima8* ata point. We take into account the fact that the color components are a function of co-ordinates (x, y) by using the notation : Ce(xy)] [RY C(x, y)=| Co (x, y)|=| G(x, y) wa (617) Coley} (Bley) | Foran image of size MXN, there are MN such vectors, C (x, y), for x=0, 1,2, =0, 1,2, N-1L .,M=Land |Color Image Processing 73 $0 now you can easily understand the figure given below. So “ “awe can apply gray scale ima; processing here also. But now this technique has to be applie “tly ns 'd to each component separately’. Mask Gray - scale Image RGB Color Image Fig. 5.12 Mask applied on color image Now after clearing these facts, now we can apply spatial filtering for color image smoothing and sharpening. 5.13.1 Color Image Smoothing Gray level smoothing can be viewed as a spatial filtering operation in which the coefficients of filtering mask are all 1’s. As the mask is slide across the image, each pixel is replaced by the average of the pixels in the neighbourhood. Thus image is smoothed. As discussed at the beginning, this concept can be extended to the full color image also means same process has to be done on each color component of color image separately. “One basic difference between gray scale image processing and color image processing is that now we have to work with vectors”. Lets,, denotes the set of co-ordinates defining a neighbourhood centered at (x, ) inan RGB color image. The average of the RGB component vectors in that neighbourhood is = 1 eae a Yew) ~+- (5.18) (yey Again in the form of vector: YR WE Sey Sc (a) w+ (5.19) (eWesy b YB) (eyesy C(x, y) = 5.3 13.2 Color-Image Sharpening ; is done by Laplacian. For whose components are he RGB colorsystem, ‘calling the sharpening of gray-scale images the color image sharpening quot analysis, we know that the Laplacian of a vector is defined as vector Walto Laplacian of individual scalar components of the input vectors. In74 Digital Image Processing VIC(x WI=]V? G(x ¥) (6.20) Vv? B(x, y) So the compute the Laplacian of a full color image, we have to compute the Laplacian of each component of image individually. 5.14 COLOR SEGMENTATION We will study segmentation in detail in a complete chapter. But here to continue the color image processing, we will study color segmentation. So for general understanding purpose, let us define segmentation. Segmentation means to divide an image in the segments. All the segments are totally separated means. One segment will not have any property that any other segmenthas. But ina single segment all the pixels will have same qualities. Segmentation of an image can be done by two methods 1. By finding similarities between pixels and arranging them ina single segment (with all simi- larities). 2. By finding differences between pixels and arranging them in different segments according to their differences. Now we will study segmentation in color images. 5.14.1 Segmentation in RGB Images ‘Suppose that, our objective is to segment objects of a specific color range in a RGB image. We have givena setof sample color points that represents the colors of our interested color range. By thatset of colors, we simply estimate the ‘average’ color that we want to segment. Let this average colorbe denoted by RGB vector ‘a’. Now our purpose is to compare the each RGB pixel of our image with this average value. One of the simplest way to find the similarities between both is Euclidean distance. Let z denote an arbitrary point in RGB space. The Euclidean distance between ‘z’ and ‘a’ is given by D(z,a)= 11 z-all = [(@-a)"(2—a)}"? = [@n—a9)* + Zq—Ac)? + (2e-49)"] where the subscripts R, G and B, denote the RGB components of vectors a and z. We say that zis similar to a if the distance between them is less than specified threshold Do. 1/2 Y summary 1. Ingray level images, image is a single valued function ;but there are other methods of imagins in which image are multi-valued function such as multispectral and multimodel. 2. Example of multispectral is color imaging. 3. By color imaging, human being easily can identify the objects.S no 10, Color Image Processing 75 Color imaging is categorized in two categories : Full color ima; with color images. In second category, converted into color images. ging where we directly deals called pseudo-color imaging, gray-level images are To describe the chromatic (color) light source ; we describe thr ee factors called radiance, lumi- nance and brightness. To differentiate different color ; we describe brightness, hue and saturation. Binary colors are red, green and blue. Secondary colors are made u ip of combining primary colors. Secondary colors are magenta, gyan and yellow. . For eyes pigment, primary colors are magenta, cyan and yellow and secondary colors are red, blueand green. . RGB, CMY (CMYK) and HSI model are some standard models for color imaging. As in case of gray-level images ; we can smooth “or’ shar; p the color images. But now same process has to be done on three planes. a Review Questions Whatare primary colors and secondary color in color imaging ? How are they defined ? » Describe the RGB color model ? How are these color can be converted to HSI color model ? . Describe HSI color model. How will you calculate hue and saturation from this model ? - What is pseudo-color-image processing ? How does a gray image convert into colorimage? What do you understand by color slicing ? . Describe @) color imaging smoothing ( color imaging sharpening. ~ Describe color image segmentation. oi