Halftoning

 

 Digital image halftoning is technique to make binary images only with black and white from color image having various lightness level and color such as scanned photos and computer graphic images[1]. Holftoned images only have black and white color. However when human eyes observe the holftoned images, they perceive continuous gradient images because human eyes spatially integrate the holtoned images in the brain operation. Generally, holftoning techniques are used to obtain high quality images from low resolution printer. In recent, due to plentiful supply of the color inkjet printers, various holftoning algorithms are researched to present previous black and white images [2].

.

 

 

Figure 1. Halftoning process from original images to binary images.

 

Figure 2. Halftoning process from original images to binary images..

 

 

Dithering

 

 Dithering for color printing is a method to express much more gray level from the limited gray level. The objective of this method removes the artificial defect caused by the resolution difference to obtain high quality images. Dithering is necessary for printers only using black and white intensity. Otherwise digital halftoning in color printing is creation of pixel pattern to make continuous colored images using limited primaries[2]. This process needs color compensation process because the images expressed by limited primaries have reduction of gray resolution and color deterioration.

 

Figure 3. Dithering for color printing.

 

 

Ordered Dither

 

 Ordered dithering determines binary pixel value from threshold matrix resulting fast process. However, for low resolution printers, deterioration is occurred as unpleasant regular pattern on printed images due to fixed threshold value. This method is classified as clustered dot ordered dither resulting dot cluster and dispersed dot ordered dither resulting dot disperse.

 

Figure 4. Single color dithering pattern..

 

 

Figure 5. Multi level dithering pattern.

 

 

 

Figure 6. Result images using ordered dithering..

 

 

Error Diffusion

 

 Error diffusion method disperses quantization error from surrounding pixels cuased by binarization of pixel value. This method has clear edge and high quality for resulting images. However, it has time consuming and worm-like pattern at bright area[4],[5]. For the application of this method to color printing, there are two representative methods; scalar error diffusion independently disperses the quantization error without considering the correlation between each channel, and vector error diffusion considering the relationship between each channel disperses the quantization error in vector color space.

 

 

Figure 7. Block diagram for error diffusion.

 

 

Figure 8. Resulting images for error diffusion.

 

 

Scalar Error Diffusion

 

 Scalar error diffusion is a independent quantization error diffusion method without consideration of relationship between each primary. This method doesnt consider the human visual system due to independent process for each channel, and has low frequency pattern as noise due to overlapped pattern. Therefore this method has lower image quality than vector error diffusion considering CMY primaries simultaneously.

 

 

Figure 9. Scalar error diffusion for a color image.

 

 

Vector Error Diffusion

 

 Vector error diffusion utilizes the CIELAB or CIEXYZ color space. It considers the correlation for each channel, then diffuse quantization error in vector color space[6]. This method gets rid of low frequency pattern noise from scalar error diffusion inducing better image quality[7]. However, this method occur unappreciated color at edges due to the large error from the quantization process. That is, large cumulated error produces other colors. Vector error diffusion uses 8 primaries (Red, Green, Blue, Cyan, Magenta, Yellow, Black, White).

 

 

 

Figure 10. Process of vector error diffusion.

 

 

Blue Noise Mask

 

 Blue noise masking(BNM) is the combination of error diffusion and ordered dithering techniques. The result pixel value can be obtained just by comparing the pixel value. At high frequency in spatial frequency, this method results flat bandwidth and binary patterns having blue noise property from 2D-mask. Blue noise which means the high frequency at spatial frequency is first called by Ulicheny. Human perceives blue noise property as good pattern. Also this noise is the opposite side of low frequency property having periodic pattern. This method has faster performance than error diffusion, however induces unclear edges and noise. The radial average power spectrum for blue noise mask is calculated as follows[9].

 

 

             (1)

 

Figure 11. Radial power spectrum for blue noise mask.

 

 

Figure 12. Resulting images for blue noise masking.

 

 

Jointly Blue Noise Mask (JBNM)

 

 Jointly blue noise mask (JBNM) is a halftoning method for color images considering pattern superposition for each color channel to remove a specific patter by human visual system. However this method occur high color difference even though reduces luminance error. Creation process for mask is ordered as creation of Initial pattern, Upward pattern, and Downward pattern[10].

 

 

ؽƮ : Figure 13. Creastion of JBNM. Progress from initial point pattern to upward pattern and downward pattern.

 

Figure 14. Progress for three initial patterns.

 

 

ؽƮ : Figure 15. Neighbor level pattern satisfying stack condition. The stack condition means next black point is determined by black point for current level. One level brighter pattern is created by removing three black points.

 

 

Modified-Jointly Blue Noise Mask (MJBNM)

 

  Modified jointly blue noise mask (MJBNM) is a halftoning method considering LPF error and S-CIELAB color difference, simultaneously. MJBNM introduce high quality blue noise pattern using the relationship between each color channel. Also, it induce small chrominance difference resulting good printed images by human visual system[11]. It uses pattern correlation for each color channel. Color difference is measured using S-CIELAB[12].

 

 

Figure 16. Measurement of S-CIELAB color difference.

 

 

JBNM                                        MJBNM

 

ؽƮ : Figure 17. Results of holftoning images for Macbeth Color Checker. MJBNM presents small luminance and chrominance difference. Especially, the color of yellow and red is express well. Also MJBNM has smooth gradient.
 

 

JBNM                                     MJBNM

 

ؽƮ : Figure 18. Comparison of halftoning result for bike image. MJBNM has better quality with smooth color expression shown in the curtain and clock.

 

 

Direct Binary Search (DBS)

 

 Direct binary search(DBS) is a halftoning method based on searching algorithm. This method has better quality compared with Screening and error diffusion method. Error diffusion can have some artifacts, however DBS induce best quality using toggle and swap [13], [14]. However, DBS needs lots of time resulting 10 times of time consumption for error diffusion. Also, this is slow for printing application. The quality of image and time are not depend on the initial pattern[15].

 

 

 

Figure 19. DBS algorithm..

 

 

Figure 20. Structure of DBS.

 

 Original image                                  DBS halfton

 

Figure 21. Resulting images using DBS algorithm.

 

 

Six color halftoning

 

Cyan, Magenta, Yellow, Black, Light Cyan, Light Magenta

 (Epson, Roland, ....)

 

Cyan, Magenta, Yellow, Black, Orange and Green

 (Pantone Hexachrome, Roland, ....)

 

 

 

Figure 22. Example of 6 color halftoning (Light magenta).

 

 

Figure 23. Comparison with color holftoning[16].

 

 

 

Six color separation to reduce graininess in a middle tone region

 

This method is a six color separation using a value of graininess pattern based on lightness and quantitative analysis to improve the graininess pattern from mixing two inks which have difference density. The value of graininess pattern is calculated using the lightness in S-CIELAB space and standard deviation for chrominance, then the graininess value and lightness value are applied to six color separation. This method has smooth and continuous gradient in a middle tone region by reducing graininess pattern property which is sensitive by human eyes.

 

 

Figure25. Algorithm for calculating the value of graininess pattern property

 


Figure 25. Six color separation using the color difference.

 


Figure 26. Six color separation using proposed lightness and the value of graininess pattern.

 

 

 

Nonlinear quantization based on printer characteristics for color dithering

 Previous quantization process for printing only considered the number of point which are colored, then the gray level of input image is separated according to the spatial property, called as linear quantization method. However, linear quantization method didnt consider the printer characteristic, exact lightness and chrominance couldnt be expressed. Therefore we proposed nonlinear quantization method considering the phenomenon which shows the overlapped areas when dyes for color printing are printed on a paper. Proposed method control the quantization level linear to the real dyed area which are remained by subtracting the overlapped area from total neighbor dyed areas.

 

 

 

Dot gain compensation for digital printer

 

 The absorption ratio of medium is depended on the property of medium. Also this is depending on the skin condition of medium coating for the same medium. Papers are coated using white mud and plastic to preserve high reflexibility and prevent creases from the painting. Especially, plastic coating has double film structure to enhance inner reflexibility resulting increase of the reflection of light. In the case of printing on papers having different ratio of absorption, the sprayed ink dots on paper have different size depending on the ratio of absorption. The increase of dot size is called as dot gain inducing large color difference on printed images. We proposed a compensation method to reduce the dot gain from different mediums based on the saturation, in addition, brightness compensation method is proposed to enhance the brightness caused by dot gain compensation.

 

 

 

Photo Text Segmentation

 

Photo test segmentation method results clearer test images by finding the text area from input scanned images. Text areas are considered having biggest Maximum gradient difference (MGD) value. In this case, texts are arrayed on the horizontal line, and make stroke upward and downward[17]. The text areas are printed through the sharpening process to emphasize[18]. Error diffusion is applied to other areas. The edge is emphasized in phased not to occur some artifacts between background and text.

 

Figure 24. Previous error diffusion method and text-emphasized error diffusion method.

 

 

References

 

[1] R. Ulichney, Digital Halftoning, The MIT Press, 1993.

[2] H. R. Kang, Digital Color Halftoning, The SPIE Optical Engineering Press, 1999.

[3] B. E. Bayer, An optimum method for two-level rendition of continuous-tone pictures, IEEE International Conference on Communications, vol. 1, pp. 26-11 to 26-15, 1976.

[4] K. T. Knox, Evolution of Error Diffusion, SPIE Conf. On Device-Independent Color Imaging, vol. 3648, pp. 448-458, Jan. 1999.

[5] R. Floyd and L. Steinberg, An adaptive algorithm for spatial gray scale, SID 1975 Symp. Dig. Tech. Papers, pp. 36-37, 1975.

[6] M. Kouzaki, T. Itoh, T. Kawaguchi, N. Tsumura, H. Haneishi, and Y. Miyake, Evaluation of digital halftone image by vector error diffusion, Proc. SPIE, 3648.

[7] S. C. Lee, Y. T. Kim, Y. H. Jo, and Y. H. Ha, "Improved Vector Error Diffusion for Reduction of Smear Artifact in the Boundary Regions," Electronic Imaging 2003, San Jose, U.S.A., vol. 5008, pp. 455-466, Jan. 2003.

[8] T. Mitsa and K. J. Parker, Digital Halftoning Technique Using a Blue-Noise Mask, J. Opt. Soc. Am. A, vol. 9, no. 11, pp 1920-1929, Nov. 1992

[9] Y. T. Kim, J. Y. Kim, H. S. Kim, and Y. H. Ha, Halftoning Method by CMY Printing Based on BNM, Eighth Color Imaging Conference: Color Science and Engineering, Scottsdale, U.S.A., pp. 252-256, Nov. 2000.

[10] M. Wang and K. J. Parker, Properties of Jointly-Blue Noise Masks and Applications to Color Halftoning, Journal of Imaging Science and Technology, vol. 44. no. 4, pp. 360-370, Jul./Aug. 2000.

[11] Y. S. Kwon, Y. T. Kim, H. K. Lee, and Y. H. Ha, "Modified Jointly Blue Noise Mask Approach Using S-CIELAB Color Difference," Journal of Imaging Science and Technology, vol. 46, no. 6, pp. 543-549, Nov./Dec. 2002.

[12] M. D. Fairchild, Color Appearance Models, Addison Wesley, 1998.

[13] M. Analoui and J. P. Allebach, Model-based halftoning using direct binary search, Proc. SPIE, vol. 1666, pp. 96-108, Feb. 1992.

[14] F. A. Baqai and J. P. Allebach, Printer models and the direct binary search algorithm, in Proc. 1998 IEEE Int. Conf. Acoustics, Speech, and Signal Processing, Seattle, WA, May 1998, pp. V-2949-V-2952.

[15] D. Kacker, T. Camis, and J. P. Allebach, Electrophotographic process embedded in direct binary search, IEEE Trans. Image Processing, vol. 11, pp. 243-357, Mar. 2002.

[16] Hexachrome homepage (http://www.dgc-hex.co.uk/Index.html).

[17] E. K. Wong, Minya Chen, A new robust algorithm for video text segmentation, Pattern Recognition, vol. 36, pp. 1397-1406, Aug. 2002.

[18] R. Eschbach, K. T. Knox, Error-diffusion algorithm with edge enhancement, J. Opt. Soc. Am. A, vol. 8, no. 12, Dec. 1991.