Image Analysis as a Tool for Printer Characterization and Halftoning Algorithm Development (invited)

Both electrophotographic (laser) and inkjet technologies can generate artifacts that limit print quality. In this paper, we review our work with the use of image analysis to develop and parameterize printer models that form the basis for modifications to the print mechanism or rendering algorithm that can reduce these artifacts. 1. Printer technologies and artifacts that impact print quality The two mainstream technologies for desktop printing are electrophotographic (EP) and inkjet (IJ). Banding is classified as those artifacts that are due to relatively high frequency quasiperiodic fluctuations in process direction parameters. Since no variation in scan direction parameters is involved, the artifacts are constant in the scan direction. With EP (or laser) printers, banding artifacts are primarily due to fluctuations in the angular velocity of the optical photoconductor (OPC) drum. These fluctuations result in non-uniform line spacing which causes a corresponding fluctuation in developed toner on the printed page. With inkjet printers, nonconstant paper speed also results in nonuniform line spacing as the printheads move back and forth in the scan direction [1, 2, 3]. Both EP and IJ printers typically generate only a limited number of states at each output pixel. To render continuoustone images with such printers, the image must first be halftoned to create textures that locally average to the desired tone. With ideal halftoning algorithms, it is implicitly assumed that either there is no interaction between neighboring dots or if there is, it is additive. This suggests that the absorptance of a halftoned patch should increase linearly with the number of dots printed in the patch. However, real printers do not obey either of these assumptions. This phenomenon is commonly called dot gain. Dot gain can be caused by one or all of the following effects: optical (trapping of scattered light under the colorant), mechanical (spreading of colorant on the medium), and electric field (only for EP printers). If halftoning is done without accounting for these nonlinearities, the printed image will not have the appearance predicted by the halftoning algorithm. One way to mitigate these problems is to use clustered dot textures. This results in visible dot texture and loss of detail. A second approach is to use tone correction. This gives correct average tone but does not give good detail rendition. The third option is to use some model of the rendering device within the halftoning algorithm [4]. With an electrophotographic printer, it is possible to develop detailed physics-based models that capture various aspects of the EP process, which include charging, exposing, developing, transferring, fusing, and cleaning. However, the behavior of all but the first two of these phases is hard to formulate. Therefore, these physics-based models are typically very complex [5]. In general, halftoning researchers have favored simpler models to characterize dot gain and local dot interaction. In the inkjet world, besides banding and dot gain, the sources of artifacts are typically twofold: dot irregularity and dot placement error. The cause of dot irregularity may be either ink coalescence or creation of satellites. Ink coalescence can take place when adjacent nozzles are fired simultaneously. Firing the nozzles at a higher frequency than they can handle may cause puddling around the nozzle opening that can create the satellites. Multiple-pass print modes are usually employed to hide these printing artifacts [6]. Fifth IEEE Southwest Symposium on Image Analysis and Interpretation (SSIAI’02) 0-7695-1537-1/02 $17.00 © 2002 IEEE Dot placement errors are caused by misaligned nozzles in a print head. The ink drop will travel roughly perpendicular to the nozzle plate at the nozzle. However, the normals to the nozzle plate at each nozzle column have a slight inward tilt as the nozzle columns are located on the curved part of the plate. This phenomenon is sometimes referred as a toe-in or dimple effect. In addition to this systematic bias, the dot displacement fluctuates randomly, which suggests the use of a stochastic model [6]. 2. Banding reduction via closed-loop feedback control Banding artifacts have been widely studied and modeled for electrophotographic printers. To reduce banding, one can either design a better mechanical system, or compensate the source of the banding through feedback. In feedback systems for banding compensation, approaches may be categorized into two groups. In the first group, we have systems that directly compensate the line spacing error via laser beam deflection [1], or directly govern the OPC drum angular velocity via motor control [2]. In the second group, we have systems that indirectly compensate the line spacing error by varying the laser beam exposure [3]. Both types of systems require a sensor to detect the OPC drum angular velocity. In our system architecture, we mount an encoder on the OPC drum shaft. Once the OPC drum angular velocity is captured as the banding signal, the compensation algorithm is then applied based on the characterized banding. To characterize banding, we use a specially designed test pattern to produce banding, and then use image analysis as a tool to analyze it. Figure 1 shows the test pattern used for this purpose. This test pattern contains three parts. The left and right fields contain scan line marks that establish the scan line position. The center field contains a bit map of alternating 0’s and 1’s that will result in banding if there is significant OPC drum velocity fluctuation. After printing the test pattern, we extract the information from the print by scanning it on a high resolution drum scanner at 2000 dpi. Line spacing and absorptance information is then computed based on the segmented results from the scanned image. A 1-D absorptance signal is obtained by projecting the scanned center field data. Figures 2a and 2b show the signal waveforms (top) and the corresponding spectra (bottom) for the scan line spacing and absorptance, respectively, before compensating the banding source. In their spectra, we observe the same main peaks, corresponding to visible banding. With different hardware architectures for banding reduction, we develop different models for compensating the source of the banding based on the image analysis results [1, 2, 3]. Figure 2c shows the projected absorptance waveLeft Scan Line Marks 50%-Fill Pattern Right Scan Line Marks 96 pixels = 0.16 in 96 pixels = 0.16 in 600 pixels = 1 in 8 8 1 1 1 Figure 1. Layout of the test pattern used to characterize banding. Banding is measured from the center 50%-fill pattern. The registration marks on either side are used to determine scan line position. 0 200 400 600 0.6 0.8 1 1.2 1.4 600 scan lines = 1 in Pr in te r pi xe ls Line Spacing 0 20 40 60 0 0.5 1 x 10 −4