Analysis for Reduction of GSR Error and DFC in Plasma TV. Sung-Jin Kang, Sang-Chul Kim, Sung-Il Chien, Member, IEEE, and Heung-Sik Tae, Member, IEEE.
P-2-8 Adaptive Stretched-out Coding Based on Cumulative Histogram Analysis for Reduction of GSR Error and DFC in Plasma TV Sung-Jin Kang, Sang-Chul Kim, Sung-Il Chien, Member, IEEE, and Heung-Sik Tae, Member, IEEE
Abstract--Like MPD-free coding method, the proposed method can achieve the complete removal of dynamic false contour (DFC) and but it can minimize the error in gray scale rendition (GSR) by adaptively adjusting subfield weight depending on the cumulative histogram of the given image.
I. INTRODUCTION Recently, the demand for slim wide-screen display equipments with a high-definition screen is increased along with the spread of digital broadcasting and the growing popularity of digital media with high picture quality. For high picture quality in the plasma display panel (PDP), it is one of the main issues to reduce dynamic false contour (DFC) [1]. The stretched-out coding such as MPD-free coding [2] and CLEAR [3] has been proposed to remove the DFC. Such removal has been achieved by the code which does not allow non-emitting subfields between successive emitting subfields. Namely, when the number of subfield is k, the stretched-out coding has only k+1 gray-levels, thereby requiring the halftone method such as error diffusion to represent all 256 gray-levels. Especially, MPD-free coding which uses a limited number of gray levels in the address and display period separated (ADS) driving scheme can completely remove the DFC but makes much error in gray scale rendition (GSR), though the halftone method is used. Thus, we propose an adaptive stretched-out coding based on histogram analysis that reduces GSR error without DFC under the ADS driving scheme. Like MPD-free coding method, the proposed method shows the advantage of the complete removal of DFC, but it determines and uses the optimal subfield weight for the given image so as to minimize GSR error. II. PROPOSED ALGORITHM When the number of the subfield in PDP-TV is k, the number of gray-levels in the stretched-out coding will be k+1 by also considering a zero level. The remaining problem that is not mentioned in the previous methods until now is how to optimize k weights of the subfield, which corresponds to the difference two neighboring gray-level values. If a histogram of the input image is uniform, GSR error will be minimized when the weights are all the same. In this case, a set of k+1 gray-levels is determined as
⎢ 255 ⎥ Gn = ⎢ × n⎥ , ⎦ ⎣ k n = 0, 1, 2, …, k,
(1)
where k is the number of the subfield. However, since a
978-1-4244-2559-4/09/$25.00 ©2009 IEEE
histogram of an image is hardly uniform, a cumulative histogram is introduced in order to reflect upon a distribution of gray-levels. Because a cumulative histogram is singlevalued and monotonically increasing, (1) is modified into ⎛ ⎢ 255 ⎥ ⎞ Gn = H −1 ⎜⎜ ⎢ × n⎥ ⎟⎟ , ⎦⎠ ⎝⎣ k n = 0, 1, 2, …, k,
(2)
where H-1 is the inverse function of a cumulative histogram normalized up to 255. When we evaluate the rendering error of the gray-levels given by (2) compared with that of the MPD-free coding method in terms of PSNR over 25 testing images, it is found that the new gray-level system performs about 5 dB better on the average sense. However, when the qualities of the picture is inspected thoroughly, it is found that the worm-like noise patterns are sometimes observed, which is analyzed to be due to the flat part of a histogram and that the interval between two adjacent gray-levels should be smaller than a threshold to suppress such noise patterns. Further observation reveals that the allowable interval between two gray-levels is shorter in low gray-levels and longer in higher gray-levels and that this discrepancy can be more easily adjusted in the domain reflecting the human visual characteristics. Perceived luminance Pn vs Gn is described by [4] β
⎛G ⎞ Pn = c × ⎜ n ⎟ , ⎝ 255 ⎠ n = 0, 1, 2, …, k,
(3)
where c is 255 as a scale factor and β is 0.382. The variable interval in the Gn domain becomes almost the same interval in the Pn domain. Through an experiment, it is confirmed that such noise is not almost observed when the interval between two adjacent perceived luminance is shorter than about 35. Therefore, if the difference between Pn and Pn-1 is bigger than 35, the noise suppression algorithm is introduced to make their difference to be smaller than 35. The error occurring in the modification of Pn and Pn-1 is made to diffuse to the other parts of gray-levels for better picture quality. After Pn is determined, a set of gray-levels on the Gn domain is calculated by 1/ β
⎛P ⎞ ′ Gn = 255 × ⎜ n ⎟ ⎝ c ⎠ n = 0, 1, 2, …, k.
,
(4)
For a hardware implementation, (3) and (4) can be represented by the look-up table with 256 entries, respectively.
Table I PSNR AVERAGED OVER 25 SAMPLE IMAGES WITH RESPECT TO MPD-FREE CODING METHOD AND PROPOSED METHOD.
Method MPD-free coding Proposed method Input image
Inverse gamma correction Cumulative histogram calculation : H Initial gray-level calculation ⎛ 255 ⎞ × n⎟ Gn = H −1 ⎜ ⎝ k ⎠
PSNR 30.58 35.07 P0 = 0 ; n = 0
n = n +1 ⎛G ⎞ Pn = c × ⎜ n ⎟ ⎝ 255 ⎠
β
No
Pn − Pn−1 > T
(a)
Yes
Noise suppression algorithm No
n>k Yes
Optimal gray-level calculation ⎛P ⎞ ′ Gn = 255 × ⎜ n ⎟ ⎝ c ⎠
1/ β
Subfield weight calculation Grayscale rendering using halftone method
Fig. 1. Flow diagram of proposed method.
′ Now, the proposed method converts Gn into a set of subfield weights. Since the stretched-out coding does not allow non-emitting subfields between successive emitting subfields, the gray- levels used in the stretched-out coding are n ′ Gn = ∑Wi , i =1
(5)
n = 1, 2, …, k, where Wi is the weight of i-th subfield. Thus, k weights Wn is given by ′ ′ Wn = Gn − Gn−1 ,
n = 1, 2, …, k.
(6)
The flow diagram of the proposed method is summarized in Fig. 1. Table I shows the PSNRs values averaged over 25 sample images for MPD-free coding method and proposed method. The PSNR of the proposed method is about 4.49 dB higher than that of the conventional MPD-free coding method on the average. Fig. 2 shows a sample example of the experimental results for the rose image. Fig. 2(a) is the experimental result by applying the MPD-free coding method to the rose image, in which the originally uniform regions are corrupted with so many dots. In contrast, as shown in Fig. 2(b), the gray-level rendering using the proposed method reduces noise patterns and GSR errors more successfully and enhances image quality considerably.
(b) Fig. 2. Experimental results using (a) MPD-free coding method and (b) proposed method.
III. CONCLUSIONS We propose an adaptive stretched-out coding based on cumulative histogram analysis that reduces GSR error without DFC under the ADS driving scheme. Although conventional stretched-out coding methods can remove DFC of PDPs completely, much error in GSR is inevitable because 256 gray-levels are represented by a limited number of gray levels. However, the proposed method determines the subfield weight based on the analysis of cumulative histogram and human visual characteristics for the given image so as to minimize GSR error, thus changing the subfield weight of PDPs dynamically. Various experiments confirm that the proposed method considerably enhances image quality of PDP-TV along with about 4.49 dB increment of PSNR. REFERENCES [1] [2] [3] [4]
T. Yamaguchi, T. Masuda, A. Kohgami, and S. Mikoshiba, “Degradation of moving-image quality in PDPs: Dynamic false contours,” J. Society for Information Display, 4/4, pp.263–270, 1996. I. Kawahara and K. Sekimoto, “Dynamic gray-scale control to reduce motion-picture disturbance for high-resolution PDP’s,” Society for Information Display Symposium Digest, pp.166–169, 1999. T. Tokunaga, H. Nakamura, M. Suzuki, and N. Saegusa, “Development of new driving method for AC-PDPs,” IDW’99 Digest, pp.787–790, 1999. J.C. Stevens and S.S. Stevens, “Brightness function: effects of adaptation,” Opt. Soc. Am., vol. 53, no. 3, pp. 375-385, 1963.
