Multitoning Method Based on Threshold Modulation Using MJBNM for Banding Artifact Reduction

This paper proposes a multitoning method based on threshold modulation using a Modified Jointly Blue Noise Mask (MJBNM) to reduce banding artifacts. Since banding artifacts appear as uniform dot distributions around the intermediate output levels, such halftone patterns result in discontinuity and a visually unpleasing output in smooth transition regions. Therefore, to reduce these banding artifacts, the principal cause of banding artifacts is first analyzed. Based on the analytical results, the proposed method then arranges the dot distribution by introducing pixels to the neighborhood of the output levels through threshold modulation using an MJBNM, which takes into account the chrominance error and correlation between channels. Depending on the input value, the original threshold range of the MJBNM is first scaled linearly so that the minimum and maximum of the scaled range include some more pixels than the adjacent two output levels containing the input value. If an input pixel is inside the vicinity of any intermediate output level producing banding artifacts, the output is set to one of those output levels based on comparing the result to scaled threshold values and a threshold modulation parameter that determines the dot density. Otherwise, a conventional multitoning method is applied. As a result, the proposed method effectively decreases the appearance of banding artifacts around the intermediate output levels. Introduction Multitoning[1],[2] is essentially an extension of bi-level halftoning[3],[4], i.e. black and white, that introduces more intermediate levels between the ON and OFF states to produce the appearance of continuous tone images. Various multitoning techniques are already widely used for better image reproduction, and advancements in printing technologies have led to research efforts on improved multitoning algorithms. Nonetheless, although multitoning is the predominant color reproduction methods, it still suffers from banding artifacts around the intermediate output levels. These banding artifacts appear as uniform dot distributions and result in discontinuity and a visually unpleasing halftone output in smooth transition regions at the printer output level. Thus, F. Faheem et al.[5] suggested a novel method based on the idea of a gray level separation to eliminate such unwanted banding artifacts, where a generalized gray level transform is used to decompose the image into constituent gray images according to the dot growth pattern defined by the transform. Each channel is then halftoned in a correlated fashion. Although this method is simple to implement and reduces banding artifacts, a high frequency granularity is visible in the mid-tones. Also, different dot growth patterns are used for the low and high frequency regions due to the image-dependent characteristic. And Q. Yu et al.[6] proposed an over modulation method to achieve a smoother transition for the intermediate output levels based on stochastic screening. With this simple technique, the dot patterns around the intermediate output levels are manipulated to introduce the desired halftone patterns. However, since this algorithm is mean preserving with respect to the input, a preprocessing step has to be added to modify the input pixel values according to an over modulation function. Also special correlation or spectral characteristics have to be considered during the screen design, as a regular screen is not optimal for this method. Cause of Banding Artifacts in Multitoning Based on Stochastic Screening The technique of stochastic screening can easily be generalized to multitoning due to its implementation simplicity. Before an input value is compared pixel by pixel to a threshold value in the stochastic screen, the threshold range, which originally has 256 levels from 0 to 255, is scaled to a certain intermediate range. Figure 1 shows the transformation of the threshold range according to the input value. However, the multitoning result reveals that banding artifacts appear inside any neighborhood with an intermediate output level. For example, if the input value is inside [82, 85], the threshold range is scaled using the following equation: