Use of Gamma Encoder for Image Processing considering Human Visualization

Transcription

1 Use of Gamma Encoer for Image Processing consiering Human Visualization M. Zahi Hasan Lecturer, Dept. of CSE Green Universit Dhaka, Banglaesh T. M. Shahriar Sazza Universit of St Anrews UK M. Hasibur Rahman EEE, BUET Dhaka, Banglaesh ABSTRACT Image processing is a promptl eveloping fiel which fins more an more application in various information an technical sstems such as: raar-tracking, communications, televisions, Biomeical image, etc. The RGB color moel is stanar esign of computer graphics sstem is not ieal for all of its applications. The re, green an blue colors are highl correlate. This makes it ifficult to eecute the image processing algorithm. Gamma encoing of images is require to compensate for properties of human vision, to maimize the use of the bits or banwith relative to how humans perceive light an color. Human vision uner common illumination conitions follows an approimate gamma or power function. If images are not gamma encoe, the allocate too man bits or too much banwith to highlights that humans cannot ifferentiate, an too few bits/banwith to shaow values that humans are sensitive to an woul require more bits/banwith to maintain the same visual qualit. Image enhancement is another technique to improve the image qualit for human visualization but sometimes it oes not improve the qualit when the images nee to be arkene or brighten. Hence, this is not a goo iea to brighten images all the time when better human visualization can be obtaine while arkening the images. Better human visualization is important for manual image processing which leas to compare the outcome with the semi-automate or automate one. Consiering the importance of gamma encoing in image processing we propose a new metho of image analsis approach which will improve visual qualit for manual processing as well as will lea analzers to analze images automaticall for comparison an testing purpose. Kewors HSI color moel, Human Visualization, Gamma Encoer, Image processing, RGB Color moel.. INTRODUCTION Color moel sstem is use to represent color. Moreover, it is a mathematical moel which is use to escribe how colors can be represente. Color space is use to escribe how the components are to be interprete. Colors can be seen as variable combinations of primar colors. Primar colors of light are aitive an hence aitive primar colors are re, green an blue. Combinations of R+G+B creates white. Moreover, primar colors of pigment are subtractive an hence subtractive primar colors are can, magenta an ellow. Combinations of C+M+Y create black []. There eist several methos to specif a color quantitativel, among etensivel use is RGB color moel where ifferent colors are ae together in ifferent was to prouce a wie range of colors. As for eample for a 24 bit RGB color image, a total number of colors can be (2 8 ) = 6,777,26. RGB color moel is use to represent an ispla images in electronic sstems. It is to mention that RGB color moel is evice epenent as Re, Green an Blue levels are ifferent from manufacturers to manufacturers [2]. Sometimes these colors var even in same evices over a perio of time an hence without a color management RGB color value oes not acts as same in evices. To improve the qualit of visual perception for color images, the term image enhancement is an important factor []. Image enhancement is neee in man areas such as photograph, scanning, image analsis etc. Image enhancement approaches fall into two broa categories such as spatial omain an frequenc omain methos [4]. The term spatial omain refers to the image plane itself, an approaches in this categor are base on irect manipulation of piels in an image whereas frequenc omain processing techniques are base on moifing the Fourier transform of an image. Images can be gra-level images or color images. Comparing with color images gra-level images have got onl one value for each piel as images are mae with piel representation. There are man eisting algorithm available which helps to enhance the image contrast for gra-level images consiering piecewise-linear transformation function name contrast stretching with normalization, stretching with histogram techniques. Most of these available algorithm are not suitable for color images although the are use wiel having poor qualit an istorte effects [5]. Gra level transformation is prove to be better approach than an other transformation an hence most propose methos are base on spatial omain approach. Image enhancement using spatial omain works with gra-level transformation or power law transformation. Power law equation is referre to as gamma. S = Cr ɤ where c an r are positive constants. Value of c= an the value of gamma can var to set the esire result an the process use to correct power-law transformation phenomena is calle gamma correction or gamma encoing. However, it is to mention that, onl enhancing the image oes not improve the image qualit for better visual perception. Sometimes it is neee to arken the bright images to obtain a better visualization [6]. Gamma is one of the main factor which helps to brighten or arken an image. 2. METHODOLOGY We have propose gamma encoing technique using spatial omain instea of frequenc omain approach. Again, as mentione earlier in RGB, there are three primar colors consiere name Re, Green an Blue where RGB is efine as aitive or subtractive moel an hence ifferent colors can be preforme using the combination of these primar colors. The RGB color moel is stanar esign of computer

2 graphics sstems not ieal for all of its applications. The re, green, an blue color components are highl correlate. This makes it ifficult to eecute the image processing algorithms. Man processing techniques work on the intensit component of an image onl. These processes are stanar implemente using the HSI color moel. In HSI color moel, color in ecompose in hue, saturation an intensit value an thus eas for human visualization. The HSI moel escribes more eact color than RGB moel escribes for human interpretation [7]. Hue is the main attribute of a color an thus ecies which color the piel has obtaine. However, hue shoul not be change at an point because changing the hue changes the color as well as istortion occurs in the image. Moreover, comparing with color space like CIE LUV an CIE Lab, in HSB it is eas to control hue an color shifting. Our main approach is to preserve the hue an appl better human visualization using saturation an brightness an hence we have chosen HSI color space instea of other color space. As mentione earlier that image are prepare in the meical laborator are RGB images. It is important to convert the RGB images into HSI images so that we can have hue, saturation an intensit in ifferentl. Our main goal is to change the properties of Saturation an Intensit an preserve the hue, so we have chosen the HSI color moel for better human visualization instea of choosing other color moel.. COLOR MODEL CONVERSION. RGB to HSI Equation () escribes the conversion from RGB to HSI color space. I S ( R G B) () min( R, G, B) ( R G B) (2) 0.5 (( R _ G) ( R B)) H cos () 2 ( R G) ( R B)( G B) If B is greater than G, then H=60 o -H (4) Where R, G an B are three color component of source RGB image, H, S an I it s components of harware inepenent on HSI format [8]..2 HSI to RGB As it can be seen that conversion from RGB to HSI is not eas with regar to computing algorithm compleit because it's regaring minimum from three searching (epression, as minimum two operators of conition), long cosine function, square root, square computation, aitional operation of conition (epression 4) uring one piel conversion. More ifficult to convert from HSI color space back to stanar RGB, where the process epens on which color sector H lies in. For the RG sector (0 0 H 20 0 ), we have the following equations to convert RGB to HSI format: B=I(-S) (5) S cos H I 0 cos(60 H) R (6) G = I (R + B) (7) For the GB sector (20 0 H ): H = H 20 0 (8) R = I ( S) (9) S cos H I 0 cos(60 H) G (0) B = I (R +G) () For the BR sector ( ): H = H (2) G = I ( S) () S cos H B I o cos(60 H) (4) R = I (G + B) (5) Fig : Clinrical Color Space of HSI format 4. GAMMA ENCODER It is wise to use luma which represents the brightness in an image an can be enote as Y. Luma is weighte average of gamma-encoing which can be enote as Y for R,G an B an hence enote as R G B. The equation becomes, Y=0.226R+0.752G B Y =0.226R G B for luminance for gamma encoing For better human visualization, the contrast enhancement operation base on the manipulation of the image histogram is histogram equalization. Initiall, we will assume a gre-scale input image, enote I input (X) If the variable is continuous an normalize to lie within the range [0,], then this allows us to consier the normalize image histogram as a probabilit ensit function (PDF) p (X), which efines the likelihoo of given gre-scale values occurring within the vicinit of. Similarl, we can enote the resulting gre-scale output image after histogram equalization as I output (X) with corresponing PDF P(). 2

3 A stanar result from elementar probabilit theor states that: p ) ( ) P ( (6) which implies that the esire output PDF epens onl on the known input PDF an the transformation function =f(). Consier, then, the following transformation function, which calculates the area uner the input probabilit ensit curve (i.e. integral) between 0 an an upper limit : ' ' ( ) p ( ) 0 (7) Differentiating this formula, appling Leibniz s rule an substituting into our previous statement we obtain the following: p ( ) p ( ) (8) p ( ) Finall, because p () is a probabilit ensit an guarantee to be positive (0 p () ). we can thus obtain []: p ( ) p ( ) 0 (9) p ( ) 5. PROCESSING STEPS This eperiment is ivie into following steps consierations for better human visualization. ) Selection of a color image in RGB format. 2) Get the values (r,g,b) for each piel for that specific image. ) Conversion of RGB color image to HSI color image. 4) Gamma encoing applie for brightness or arkness for better visualization. 5) Saturation value applie using histogram equalization. 6) Conversion of HSI color image to RGB color image. 7) Save an use the resultant image for other image analsis. 6. EXPERIMENTAL RESULTS Histogram Equalization metho applie to the original color images where this metho changes the color value (hue) of the original images. It is known that majorit methos of image processing working onl with intensit part of color moel [9- ]. The color moel must be in full basis, it mean that moel must allow to transform image to new color moel, use the intensit component for image processing then return image back to RGB after processing [2,]. RGB Image RGB to HSI Conversion Gamma encoing for brightness an arkness HSI to RGB Conversion RGB Image Saturation Fig : Sstem Block Diagram To evaluate the performance of our propose metho, Gamma encoing helps to maintain the visual qualit of images. To evaluate the contrast performance we have applie histogram equalization saturation value from an gamma correction value ranges from in ifferent computers as ifferent computers acts ifferent accoring to gamma value. It is to mention that gamma value > performs arkening an vice-versa. In this section we present a hue preserving gamma encoing metho base on the HSI color space. A comparison among our propose metho an the image enhancement metho is carrie out as shown in Fig. (-). Fig. shows the results for the first image, John Coltrane where the image enhancement technique generate a arken image (Fig. (b)). In turn, the image prouce b our propose metho (Fig. (c)) is more realistic than the others. We can sa that the resulte image of gamma encoing metho has better qualit than the others. Fig. 2(b) shows the results for the secon image, the original image prouces an over enhance image, that is, the colors are ver saturate. In Fig. 2(c) that our metho generates an image with a goo balance between non saturate an realistic colors. From the iscussion above, we claim that our metho prouces images (Fig ((c)-(c))) with the best traeoff

4 Fig (a): Original Image Fig (b): Histogram Equalization Fig (c): Propose Metho Fig 2(a): Original Image Fig 2(b): Histogram Equalization Fig 2(c): Propose Metho Fig (a): Original Image Fig (b): Histogram Equalization Fig (c): Propose Metho Table : Comparison of eisting an propose metho with accurac Metho Use Number of Images Error (%) Accurac (%) Eisting Metho 20 26% 74% Propose Metho 20 % 87% between the enhance colors an saturation. That is, our metho prouces images with colors that are more realistic than the image enhancement technique (which are not hue preserving), an the images are not as saturate as the ones prouce b the metho. 7. CONCLUSION This paper has propose a color enhancement approach using luminance component on gamma correction base on human visualization as well as saturation component. The software was implemente using MATLAB. As shown on the eperiments in the previous section, it is ifficult to juge an enhance image result even with a subjective assessment. However, we claim that our metho is more robust than the others in the sense that neither unrealistic colors nor over 4

5 enhance are prouce. For future works, we plan to evaluate the methos using naturalness an colorfulness metrics on a atabase with hunres of images collecte from internet, such that a quantitative comparison can be performe. However, there ma be still some areas nees to be taken care of as the color enhancement nees to change or shift color using hue although these cases are eceptional an ver rare. 8. REFERENCES [] C. Solomon, T. Breckon. 20. Funamentals of Digital Image Processing. [2] Nishu, Sunil A. 202, Quantifing the efect visibilit in igital images b proper color space selection, International journal of engineering research an applications, vol.2, Issue, pp [] Raunaq M. an Utkarsh U., 2008, Hus-preserving color image enhancement without gamut problem, Term paper, pp. -6. [4] Yusuf Abu S., Nija Al-Najawi, Sara T., 20, Eploiting Hbri methos for enhancing igital X-Ra Images, International Arab journal of information technolog, vol. 8. [5] Umesh R., Zhou W., Eero P. S., 2009, Quantifing color image istortions base on aaptive spatio-chromatic signal ecompositions, IEEE international conference on image processing. [6] Hana Al-Nuaim, Nouf A., 20, A user perceive qualit assessment of loss compresse images, International journal of computer graphics, vol. 2, No. 2, pp [7] R.C. Gonzalez an R.E. woos, 2007, Digital Image Processing, r Eition, Prentice Hall, Upper Sale River, NJ. [8] Jian-feng Li, Kaun-Quan Wang, Davi Zhang, 2002, A New equation of saturation in RGB-TO-HIS conversion for more rapiit of computing, Proceeings of the international conference on machine learning an cbernertics, pp [9] Papoulis, A., 968, Sstems an Transforms with Applications in Optics, New York: McGraw-Hill. [0] Russ, J.C., 995, the Image Processing Hanbook. Secon e., Boca Raton, Floria: CRC Press. [] R. E. Blake, 999, Partitioning Graph Matching with Constraints, Pattern Recognition, Vol 27, No., pp [2] J. Fole, A. van Dam, S. Feiner an J. Hughes, 990, Computer Graphics: Principles an Practice, Secon Eition, Aison-Wesle, Reaing, MA. [] R. E. Blake an P. Boros, 995, The Etraction of Structural Features for Use in Computer Vision. Proceeings of the Secon Asian Conference on Computer Vision, Singapore. 5