JP2006031003A - Method and system for improving display resolution using subpixel sampling and visual error compensation - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for suitably converting a high-resolution image into a low-resolution image by using visible errors. <P>SOLUTION: The method comprises the following steps. Sub-pixel sampling simulation (72) is performed on a high-resolution image (70) to determine an error to be introduced into the high-resolution image as a result of sub-pixel sampling. The error may be isolated from the high-resolution image to create an error image. The error image may be modified with a visual model (76) to remove invisible errors, and thereby a visible error image is created. Then the visual error image may be combined with the high-resolution image to create a compensated image (82), and when sub-pixel sampling (80) occurs, the compensated image substantially cancels the introduced error as a result of subsequent sub-pixel sampling. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

(Refer to related applications)
This application is based on Scott J. et al. It is a continuation of US patent application serial number 09 / 735,454 filed December 12, 2000 with the invention by Daly, which is also incorporated by Dean Messaging, Scott J. et al. It is a continuation of US patent application serial number 10 / 447,186, filed May 27, 2003, invented by Dary, Rajesh Reddy Kovvuri, both of which are US provisional patents filed June 12, 2000 Claim the benefit of application number 60 / 211,020.

  Embodiments of the invention relate to the field of displaying high resolution images on a display using a lower resolution. In this case, the display uses a triad to display the R, G, B component of the image. This triad is common, for example, in direct view LCD displays, and in such an arrangement, a single pixel constitutes three adjacent subpixels. Each sub-pixel controls only one of the three primary colors (ie, R, G, B) and is typically controlled solely by the primary color of the digital image display. High resolution images may be available in storage or directly from algorithms (vector graphics, some font designs, computer graphics).

  The most common method used to display high resolution on a lower resolution display is to sample the high resolution 4 pixels 2 to the resolution of the low resolution display 6, as shown in FIG. The downsampled values R, G, B of each color pixel 8 are then mapped to separate R, G, B elements 10, 12, 14 of each display pixel 16. These R, G, B elements 10, 12, 14 of the display pixel are also called subpixels. Since the display device cannot overlap the color elements, only the sub-pixel can acquire one of the three R, G, B colors, but the color amplitude is in the entire gray scale range (eg, 0-255). Change through. Subpixels typically have an aspect ratio (width: height) of 1: 3, resulting in pixels 16 being square. Subsampling / mapping techniques do not take into account the fact that the R, G, B subpixels of the display are spatially displaced. In fact, those subpixels are considered overlapping as in the high resolution image. This type of sampling may be referred to as sub-sampling or conventional sub-sampling.

  The pixels of the high resolution image 4 are shown as three slightly offset stacked squares 8, indicating that the RGB values are associated with the same spatial location (ie, pixel). One display pixel 16, each consisting of one R, G, B subpixel 10, 12, 14, is shown as part of the lower resolution triad display 6 of FIG. 1 using dark lines. Other display pixels are shown in lighter gray lines.

  In this example, the high resolution image has three times more resolution than the display (in both parallel and vertical directions). Since this direct sub-sampling technique causes aliasing artifacts, various methods are used, for example, using sampled pixels to average adjacent unsampled pixels. It should be noted that the general technique of averaging adjacent elements while subsampling is mathematically equivalent to pre-filtering a high resolution image using a rect filter. . It should also be noted that the technique of selecting a pixel different from the leftmost pixel (shown in this figure) is considered pre-filtering that only affects the phase. Thus, most of the processing related to anti-aliasing is considered a high resolution filtering operation even if the kernel is applied only to the sampled pixel location.

  It will be appreciated that the foregoing techniques do not take advantage of potential display resolution. For background information on this field, see R.A. Fiegenblatt (1989) “Full color imaging on amplicolor color displays” Proc. SPIE V. 1075, 19-205; Kranz and L.L. Silverstein (1990) "Color matrix display image quality: The effects of luminance and spatial sampling," SID Symp. Can be obtained by referring to Digest 29-32. These two books are hereby incorporated by reference.

  For example, in the display shown in FIG. 1, the display pixel 16 resolution is one third of the resolution of the high resolution image (source image) 4, while the resolution of the subpixels 10, 12 and 14 is the source (horizontal). Equal to the resolution of the sub-pixel. If this display is used only for those who are insensitive to color, it is possible to take advantage of the spatial location of the subpixels. This approach is illustrated in FIG. 2 below. Here, the R, G, B subpixels 10, 12, 14 of the display are taken from the corresponding colors of the different pixels 11, 13, 15 of the high resolution image. This allows the horizontal resolution to be a sub-pixel resolution that is three times the display pixel resolution.

  But what about display viewers who are not insensitive to colors? That is, the majority of viewers. Fortunately, display engineers are insensitive to color at the highest spatial frequencies, even for observers with full color vision. This is shown in FIG. 3 below. Here, the ideal spatial frequency response of the human visual system is shown.

  Here, luminance 17 is associated with achromatic contact of the displayed image and chrominance 19 is associated with color content. The color content is processed by the visual system as an equivalent glow from red to green and blue to yellow. The image color difference signals RY and BY are approximately equal to these modulations. For most observers, the bandwidth of the color frequency response is half that of the luminance frequency response. Sometimes the bandwidth of the blue-yellow modulation response becomes narrower and decreases to approximately one third of the luminance. Sampling that constitutes the mapping of display pixel triad color elements from different image pixels may be referred to as sub-pixel sampling.

  Referring to FIG. 4, in the horizontal direction of the display, the Nyquist of display pixel 16 (display pixel = triad pixel, providing triad Nyquist at 0.5 cycles per triad pixel) and Nyquist of subpixel elements 10, 12, 14 There is a frequency range between the frequencies (0.5 cycles per subpixel = 1.5 cycles per triad pixel). This portion is shown as a rectangular portion 20 in FIG. The sinc function resulting from wrapping a high resolution image with a rect function having a width equal to the display sample spacing is shown as a thin dashed-dot curve 22. This is the most common approach taken to modulate the display MTF (Modulation Transfer Function) when the display is an LCD.

  The sinc function that results from wrapping the high resolution source image with rect equal to subpixel spacing is shown as a dashed curve 24. This has a wider bandwidth. This is a limitation imposed by the display, with the idea that subpixels are rect in 1D. In the rectangular portion 20 shown, the sub-pixel can display luminance information but not color information. In fact, the color information in this part is aliased. Thus, by allowing color aliasing in this part, higher frequency luminance information can be achieved than triad (ie, display) pixels. This is an “advantageous” part of using subpixel sampling.

  In applications with a font display, black and white fonts are typically preprocessed, as shown in FIG. Standard preprocessing includes hinting. Hinting is related to the centering of the font stroke at the center of the pixel. That is, a font stroke specific phase shift. This usually occurs following low pass filtering. Low pass filtering is also referred to as grayscale anti-aliasing.

  The visual frequency response (CSF) shown in FIG. 3 is idealized. In practice, it has a finite falloff gradient as shown in FIG. 6A. Luminance CSF 30 is mapped from cy / deg units to the display pixel domain (assuming a viewing distance of 1280 pixels). This is shown as a solid line 30 with a maximum frequency near 1.5 cy per pixel (display pixel) and is a band pass with a peak near 2 cy per pixel triad. R: G CSF32 is shown as a dashed line. This is a low pass with a maximum frequency near 0.5 cy per pixel. B: Y modulated CSF 34 is shown as a dashed-dot LPF curve with the same maximum frequency as R: G CSF, but with a lower maximum response. The range between the cutoff frequency of the chromaticity CSFs 32 and 34 and the luminance CSF 30 is the portion that allows color aliasing to improve the luminance bandwidth.

  FIG. 6A also shows the idealized image power spectrum 36 as a 1 / f function, represented in the figure as a straight line with a slope of −1 (since the figure uses a logarithmic axis). This range is repeated at the sampling frequency. These iterations show the sampling rate of horizontal pixels 38 and sub-pixels 40. Generation at the lower frequency 38 occurs by pixel sampling, and generation at the higher frequency 40 occurs by subpixel sampling. It should be noted that the shape changes because it is plotted on the logarithmic frequency axis. The frequency of these repeat ranges extending to lower frequencies below Nyquist is called aliasing. The leftmost color aliasing 38 is due to the pixel sampling ratio. On the other hand, luminance aliasing 40 occurs at higher frequencies as it relates to higher subpixel sampling ratios.

  In FIG. 6A, no pre-filtering is applied to the source range. Therefore, aliasing due to pixel sampling (ie color aliasing) extends to the low frequency 35. Thus, even though the color CSF has a narrower bandwidth than the luminance CSF, color artifacts can still be visible (due to noise and display contrast).

  In FIG. 6B, the prefilter (rect function equal to three source image pixels), shown in FIG. 4 as dashed-dotted line 22, was applied to the source power spectrum. It then appears to affect the baseband range 42 past 0.5 cy per pixel, causing a steeper slope than the −1 seen at 44. The iteration also shows the effect of this prefilter. Even with this filter, it can be seen that some color aliasing (repeated range at lower frequencies) occurs at a frequency 46 that is lower than the cutoff frequency of the two colors CSF 32a and 34a. Thus, it can be seen that simple luminance pre-filtering is difficult to remove color aliasing unless all luminance frequencies past 0.5 cy per pixel (ie, the “favorable” portion) are removed. Will.

  One possibility is to design the pre-filtering based on the visual system model, as it depends on the visual system difference in bandwidth as a function of luminance or color that expands the luminance bandwidth in the “advantageous” part 20 It is. This is incorporated herein by reference and is shown in FIG. Betrisey et al. (2000) "Distributed filtering for patterned displays", SID Symposium digests, 296-299.

  This technique ideally uses different filters depending on the type of color layer and the color subpixel from which the image is sampled. That is, there are nine filters. It was designed using a human visual difference model. This is incorporated herein and described in the following work, shown in FIG. Zhang and B.H. Wandell (1996) "A spatial extension of CIELAB for digital color image re- production", SID Symp. Digest 731-734. This was done offline, assuming that the images were always black and white. In the final execution stage, the rect function is used instead of the resulting filter to save the calculation. Furthermore, some residual color errors are still seen because color aliasing extends to a lower frequency than color CSF cutoff (as seen in FIG. 6B).

  However, the visual model used does not take into account the masking of visual system properties that cause color masking by luminance when the luminance is at a medium to high contrast level. Therefore, larger font color artifacts placed along the edge of the font are masked by the high brightness contrast of the font. However, when the font size is reduced, the font brightness decreases, and the same color artifact becomes very visible (for example, even for very small fonts, the b / w portion of the font disappears, only localized color speckles Is left).

  Embodiments of the present invention constitute methods and systems that are largely independent of filtering and its linearity assumptions. These embodiments take into account specific localized phases of various font strokes or other image characteristics. The method of embodiments of the present invention is based on compensation, where a signal is added to the image and the image nullifies color errors caused by sub-pixel addressing. In some embodiments, only lower frequency color artifacts are disabled. This is because, as shown in FIG. 6A, high frequency artifacts are not visible due to the lower bandwidth of color CSF.

  Embodiments of the present invention provide a higher resolution luminance signal, especially when the display is not viewed closer to the designed specification, and the visibility of color aliasing is reduced. Furthermore, the model of the present invention is not limited to the use of black and white source images or text documents. These methods usually increase the horizontal resolution of the color image. Also, by using more advanced embodiments, the visual effect of human masking is taken into account, and more color aliasing near the high contrast luminance edge can be used to further enhance the higher luminance edge.

  Embodiments of the present invention transform a high resolution image into a lower resolution image by an error reduction method that yields a lower resolution image that more accurately represents the original high resolution image. High resolution images can be sampled in a sampling simulation to simulate visible errors caused by subpixel sampling. Since sampling is only simulated, the resolution need not change. When an error is determined, the error is subtracted from the original image to create an error image. A visual model can then be used to remove invisible detectable information from the error image, creating a modified error image that represents the visible error.

The compensation image can then be added to the original image so that the error is added to the original image. The error is subsequently removed during subpixel sampling. The compensated image is then sampled using a subpixel sampling process. The sub-pixel sampling process removes errors added by the results of the process and the improved lower resolution image.
(Item 1)
A method of resampling an image,
a) providing a channel of the image comprising at least 60% of the brightness of the image;
b) attenuating at least a portion of the lower frequency luminance information relative to at least a portion of the higher frequency luminance information of the channel;
c) resampling the luminance information of the image.

(Item 2)
The method of claim 1, further comprising attenuating at least a portion of the lower frequency color information relative to at least a portion of the higher frequency color information as a result of the resampling of the luminance information.

(Item 3)
A method of resampling an image,
(A) re-sampling the luminance information of the image, the luminance information being at least 60% of the luminance of the image;
And (b) attenuating at least a portion of the lower frequency color information with respect to at least a portion of the higher frequency color information as a result of the resampling of the luminance information.

(Item 4)
4. The method of item 3, wherein resampling the luminance information includes pixel resampling.

(Item 5)
4. A method according to item 3, wherein the attenuating is using a high-pass filter.

(Item 6)
4. The method of item 3, wherein re-sampling the luminance information results in two color channels, each of which decays in a different manner.

(Item 7)
A method of resampling an image with a plurality of channels, wherein a first channel of the plurality of channels has a luminance component comprising at least 60% of the luminance of the image, and a second of the plurality of channels. Channel has a color component and the method includes:
(A) resampling the first channel of the image;
(B) attenuating at least a portion of the lower frequency color information relative to at least a portion of the higher frequency color information as a result of resampling of the first channel of the image;
(C) resampling the second channel of the image.

(Item 8)
8. A method according to item 7, wherein the luminance component comprises at least 70% of the luminance of the image.

(Item 9)
Item 8. The method of item 7, wherein the luminance component comprises at least 80% of the luminance of the image.

(Item 10)
8. The method of item 7, wherein the luminance component comprises at least 90% of the luminance of the image.

(Item 11)
Item 8. The method according to Item 7, wherein the plurality of channels are channels having different colors.
(wrap up)
Embodiments of the present invention provide a system and method for converting a high resolution image to a low resolution image using reduced visual error. These systems and methods include: A subpixel sampling simulation is performed on a high resolution image that determines errors introduced into the high resolution image as a result of subpixel sampling. This error can be isolated from the high resolution image to create an error image, which can be corrected using a visual model to remove the invisible error, creating a visible error image. The visual error image can then be combined with the high resolution image to create a compensated image, and when subpixel sampling occurs, the compensated image substantially cancels the introduced error as a result of subsequent subpixel sampling. .

  Preferred embodiments of the present invention are best understood by referring to the figures, in which like numerals indicate like parts.

  As generally illustrated herein, it is readily understood that the components of the present invention can be arranged and designed in a variety of different configurations. Thus, the following details of the method and system embodiments of the present invention are not intended to limit the scope of the invention, but are merely representative of the presently preferred embodiments of the invention.

  Elements of embodiments of the present invention may be embodied in hardware, firmware and / or software. The exemplary embodiments disclosed herein may only describe one of these forms, and those skilled in the art will recognize these elements in any of these forms, while remaining within the scope of the invention. It should be understood that could be achieved.

  Some embodiments of the present invention are summarized in the block diagram shown in FIG. Here, a high resolution image such as the RGB high resolution image 70 is corrected. The high resolution image 70 is sampled in a subpixel sampling simulation 72 to simulate the visible error caused by subpixel sampling. Since sampling is only simulated, the resolution does not change. If an error is confirmed, the error is subtracted from the original image 70 (74). The visual model 76 then removes undetectable information from the error image that creates a modified error image that represents a visible error. This corrected error image may be referred to as a compensated image and is not limited to the 0-255R, G, B range of the source image 70. This is because it allows the use of negative values.

  The compensated image is then added to the original image 70, thereby adding errors to the original image 70 that are removed during subpixel sampling. The compensated image 82 is then sampled using a subpixel sampling process 80. In the sub-pixel sampling process 80, the added error is removed as a result of the process and the improved lower resolution image.

  The main aspect is that the visible error caused by subpixel sampling is simulated and then separated from the high resolution source image. The simulation is maintained at the same resolution as the higher resolution source image. If such an error is added by the subpixel sampling process, it is invalidated in the final image.

  The method of embodiments of the present invention negates color aliasing that occurs at lower frequencies. The color aliasing frequency is originally very high, so it folds over to a very low frequency with Nyquist. Color aliasing frequency is one of the main concerns because it is the most visible.

  The visual model 76 of an embodiment of the present invention is a different form than the well-known model used in the Betrisey approach. In a well-known model such as Betrisey, the visual model is a different distance function where two images are input and the output is an image showing where the visual difference occurs. In the Betrisey approach, this visible difference image is integrated in the form of a square to arrive at a single distance function. These well-known models are described next; Zhang and B Wandell (1996), "A spatial extension of CIELAB for digital color image re- production", SID Symposium Digest 731-734; Batrisey et al. (2000), "Distributed filtering for patterned displays", SID Symposium Digest, 296-299; Dally (1993), “Visible Differences Predictor”, Ch. 14 of Digital Images and Human Vision, A.M. B. Watson (eds.), MIT Press. These references are incorporated herein.

  However, in embodiments of the present invention, the form of visual model 76 removes image content that is not visible to the eye. Thus, the visual model does not calculate the visible difference between images, but rather operates on a single image. One way to achieve this is filtering with appropriate CSF and coring with thresholds. In FIG. 8, the visual model 76 is shown in a general form.

  Referring to FIG. 9, the method of the present invention is used in conjunction with standard preprocessing methods 84 such as more conventional techniques such as hinting and low pass filters. When the standard preprocessing method 84 is executed, the image is processed in a manner similar to the method without the standard preprocessing method. That is, after the high resolution image 70 has been preprocessed (84), it is sampled in a subpixel sampling simulation 72 to simulate visible errors caused by subpixel sampling. If an error is confirmed, the error is subtracted from the original image 70 (74). Then, the visual model 76 removes the non-detectable information from the error image and creates a corrected error image representing the visible error. The compensated image is then added to the original image 70 to add an error that is removed during the actual subpixel sampling 80. The compensated image 82 is then sampled using a subpixel sampling process 80. In the sub-pixel sampling process 80, the added error is removed as a result of the process and the improved lower resolution image.

  Many embodiments of the visual model 76 can be used with embodiments of the present invention. Referring to FIG. 10, a first embodiment of the present invention is described. A high resolution image 70 has been processed. The image 70 can be preprocessed 84 by the user as needed, or directly processed without preprocessing. Image 70 is passed to subpixel sampling simulation 72 to determine errors associated with subpixel sampling. The error is then subtracted 74 from the original image 70 and an error image is generated and processed via the visual model 76. In this embodiment, visual model 76 includes RGB to LAB conversion 90. The conversion 90 results in an image that is represented by three channels, thereby decoupling the luminance characteristic from the color characteristic. While several color models can be used, the CIELAB model is exemplary and in this simple embodiment an equivalent model is preferred. Simpler embodiments using linear Y, RY, BY signals are also possible.

  When the error image is converted, the luminance channel 92, color channels 94 and 96 are filtered to remove invisible errors from the error image. In the preferred embodiment, these filtering operations typically comprise filtering and spatial coring operations to remove localized frequencies that are so small that the amplitude is not visible. Different filters can be used for each channel. Filtering is generally a frequency specific operation, although some channels are not filtered because they are required in a given application. Typically, the luminance channel 92 is filtered separately from the color channels 94 and 96.

  After filtering, channels 92, 94, and 96 are combined (98) into a single LAB format image that represents the visible error associated with subpixel sampling. This LAB visible error image is subsequently reconverted (100) and returned to the RGB format visible error image 101. The RGB visible error image 101 is then combined with the original image 70 (78) to form a compensated image 82 that compensates for errors introduced through subpixel sampling. This compensated image 82 is then sampled using subpixel sampling 80. In sub-pixel sampling 80, the added visible error compensation 101 is disabled in the sampling process, producing a lower resolution image with fewer errors than an image created through simple sub-pixel sampling alone.

  In the second embodiment of the present invention, edge effects or masking and frequency effects are processed, as shown in FIG. Typically, the method of embodiments of the present invention is performed on a high resolution image, such as an RGB high resolution image 110. The pre-processing 112 can be performed before applying the method of the embodiment of the present invention, but is not required.

  The high resolution image 110 is transmitted to the subpixel sampling simulation 114. Sub-pixel sampling simulation 114 performs sampling simulation during sub-pixel sampling and determines errors introduced by converting the sampled image to the original resolution for comparison with the original image. The sampled and transformed image is compared (116) with the original image to determine errors associated with subpixel sampling. This error is subtracted from the original image to create an error image 118 that is sent to the visual model 76 for further processing.

  Within visual model 76, the error image in RGB or similar format is converted 120 to LAB or similar format, thereby separating the luminance data from the color data. After conversion to a luminance separation format such as LAB, the separate channels of luminance 122, colors 124 and 126 are filtered according to frequency as described above for other embodiments.

  After filtering 122, 124 and 126, this embodiment of the present invention takes into account the visual properties of masking, in particular the masking of colors by luminance. The masking signal is obtained from the contents of the source image 110 rather than from the error image. The source image 110 is converted (128) into a luminance separation format such as LAB from which luminance data is extracted (130). In some embodiments, only the luminance channel 136 is masked, but the color channels 134 and 132 can also be masked. Masking is performed to produce an edge effect as a pixel-wise comparison. The contrast is, for example, an amplitude differential signal, which is obtained by subtracting the average of the entire L image from the L image, and its absolute value is proportional to the contrast. A higher contrast signal level at L at a location results in more masking of L, R / G, B / Y signals at the same location. Masking is simulated by separating these signals by the mask signal output from step 130 and then coring. Coring is a process in which the signal value changes to zero when the absolute value of the signal amplitude is smaller than a specific threshold value.

  When masking occurs, the separated channels are combined at 138 and returned to a combined format such as LAB 138, and the combined format file is then converted (140) to the original image format such as RGB. Returned. The resulting image represents a visible error 142 associated with subpixel sampling.

  This resulting error image 142 is subsequently combined with the original high resolution image 144 to produce a compensated image 146. In the compensated image 146, substantially similar corrections are introduced, but the opposite is the error introduced during subpixel sampling. This compensated image 146 is actually sampled (148), resulting in a display image 150 that contains fewer errors than the directly sampled image without error correction. This is because the sampling error is invalidated by the visible error (142) introduced before the sampling 148 (144).

  Since the actual visual masking process only masks signals within a limited range or a limited spatial portion near similar mask frequencies, the second implementation of the present invention as shown in FIG. The form can only model the mask effect as it is. However, a 1 / f power (power) spectrum can be assumed for an image consisting only of edges and lines. Thus, for a particular frequency and direction, there are fewer higher frequency signal components in the same direction. Thus, this approach overestimates the masking, but as a result, more errors are generated in the compensated image, so the net effect removes more color aliasing than necessary. As a result, the brightness sharpness is smaller but still greater than the method without masking.

  FIG. 12 will be described. FIG. 12 is a third embodiment of the present invention that uses a more complete visual model that can more accurately predict masking using multiple frequency channels. Although only four channels are shown, their actual number can be large or small, and is typically bandpassed and direction limited. An exemplary channel is S.I. Daly (1993), “Visible Differences Predictor”, Ch. 14 Digital Images and Human Vision B. Watson et al. (Eds.), MIT Press; Lubin (1995), “A Visual Discrimination Model for Imaging System Design and Evaluation”, Ch. 10 Vision Models for Target Detection and Recognition Peli (ed.) World Scientific Press, which is incorporated herein by reference. This version is processing intensive, but can become more practical with increasing computing techniques.

  In the third embodiment, the high resolution image 110 may optionally be preprocessed (112) prior to the subpixel sampling simulation (114). As in the previously described embodiments, the subpixel sampling simulation 114 is used to determine the error 118 introduced from the subpixel sampling. This error image is extracted from an image that directly compares the “processed” image with the original image 110 at the original image resolution (116). Typically, lower resolution “after processed” images are increased in resolution for comparison. Once this error image 118 is acquired, the error image 118 can be processed with the visual model 151 of the present embodiment.

  As in other embodiments, the error image 118 is converted from an RGB or similar format to a luminance separation format such as LAB 152. By using this type of format, luminance and color data are separated into channels, and each channel is divided into frequency ranges using filter bank divisions 154, 156, and 158. The respective frequency range within each channel is then filtered using band scaling 160, 162, 164.

  Edge effects also occur as follows. By converting the original source image into a luminance separation format such as LAB 166 and then filter banking the luminance channel 168 from the source image 110. In general, the separated color channel 165 is not used in the masking process. In subsequent filter bank partitioning 168, the frequency range is filtered via band scaling or similar procedures, as performed on main error image channels 160, 162, 164. These signals generated via luminance channel band scaling 170 can be used to mask the various luminance and color channels 172, 174 and 176. Although the masking calculation is similar to that described with reference to FIG. 11, in the calculation of this embodiment, the masking signal from a particular frequency band only affects the corresponding frequency band of the error image.

  Once masking of each channel to its respective frequency range is complete, the channels may be combined (178) to form a standard LAB or similar format file. This combined file can then be converted back to the original image format, such as RGB180. This RGB or similar file 180 represents the visible error introduced during subpixel sampling. The visible error 180 is then subtracted from the original high resolution source file 110 (188), or otherwise considers the original high resolution source file 110, thereby generating a compensated image 182.

  When the compensated image 182 travels through the sub-pixel sampling 184, errors introduced into the sampling process are combined with the original image 110 (188), invalidating the visible error, than images directly sampled without correction. A display image 186 with fewer errors is produced. This third embodiment uses a more accurate visual model with multi-channel capability that provides chromaticity masking by luminance.

  The present invention may be embodied in other specific forms without departing from the spirit or basic characteristics of the invention. The described embodiments are to be considered in all respects illustrated and not restrictive. The scope of the invention is, therefore, indicated by the appended claims rather than by the foregoing description. All changes that come within the meaning and range of equivalency of the claims are to be embraced within their scope.

FIG. 6 illustrates conventional image sampling for a display having a triad pixel configuration. FIG. 5 illustrates sub-pixel image sampling on a display having a triad pixel configuration. Fig. 6 is a graph showing an ideal CSF mapped to a digital frequency plane. FIG. 6 is a graph illustrating an analysis of a pixel Nyquist and sub-pixel Nyquist portion showing an advantageous portion. A typical pretreatment technique is shown. FIG. 6 is a graph showing an analysis using a 1 / f power spectrum and sub-pixel sampling frequency repeated with pixel sampling. FIG. 6 is a graph illustrating an analysis using a 1 / f power spectrum repeated with pixel sampling and a sub-pixel sampling frequency improved by preprocessing. FIG. 2 is a block diagram illustrating a known use of a visual model. FIG. 3 is a block diagram illustrating a generic embodiment of the present invention. FIG. 3 is a block diagram illustrating an embodiment of the present invention that uses preprocessing. FIG. 6 is a block diagram illustrating an embodiment of the present invention using separate luminance and color channel filtering. FIG. 4 is a block diagram of an embodiment of the present invention that uses a visual model that utilizes chromaticity masking by luminance. FIG. 6 is a block diagram of an embodiment of the present invention using a visual model that utilizes luminance chromaticity masking with a more accurate multi-channel, divided frequency range visual model.

Explanation of symbols

70 High-resolution image 72 Sub-pixel sampling simulation 76 Visual model (removal of invisible errors)
80 Subpixel sampling 82 Compensated image

Claims (11)

A method of resampling an image,
a) providing a channel of the image comprising at least 60% of the brightness of the image;
b) attenuating at least a portion of the lower frequency luminance information relative to at least a portion of the higher frequency luminance information of the channel;
c) resampling the luminance information of the image.   The method of claim 1, further comprising attenuating at least a portion of the lower frequency color information relative to at least a portion of the higher frequency color information as a result of the resampling of the luminance information. A method of resampling an image,
(A) re-sampling the luminance information of the image, the luminance information being at least 60% of the luminance of the image;
(B) attenuating at least a portion of the lower frequency color information relative to at least a portion of the higher frequency color information as a result of re-sampling of the luminance information.   The method of claim 3, wherein resampling the luminance information comprises pixel resampling.   The method of claim 3, wherein the attenuating is using a high pass filter.   4. The method of claim 3, wherein re-sampling the luminance information results in two color channels, each of which attenuates in a different manner. A method of resampling an image with a plurality of channels, wherein a first channel of the plurality of channels has a luminance component comprising at least 60% of the luminance of the image, and a second of the plurality of channels. Channel has a color component and the method includes:
(A) resampling the first channel of the image;
(B) attenuating at least a portion of the lower frequency color information relative to at least a portion of the higher frequency color information as a result of resampling of the first channel of the image;
(C) resampling the second channel of the image.   The method of claim 7, wherein the luminance component comprises at least 70% of the luminance of the image.   The method of claim 7, wherein the luminance component comprises at least 80% of the luminance of the image.   The method of claim 7, wherein the luminance component comprises at least 90% of the luminance of the image.   The method of claim 7, wherein the plurality of channels are channels of different colors.

Images