Patent Application
20080123969
The present invention relates to electronic signal processing and in particular, reducing the level of noise in localized parts of a signal's bandwidth.
The following applications have been filed by the Applicant simultaneously with the present application:
IJ70US IJ71US IJ72US IJ73US IJ74US IJ75US
The disclosures of these co-pending applications are incorporated herein by reference. The above applications have been identified by their filing docket number, which will be substituted with the corresponding application number, once assigned.
The following patents or patent applications filed by the applicant or assignee of the present invention are hereby incorporated by cross-reference.
Electronic signal processing will usually induce some level of noise into the output signal. In the transmission of signals, the induced noise is often due to ‘lossy’ transmission methods. Ordinary workers in this field will understand that ‘lossy’ refers to processing techniques that move some signal data to nearby data values.
Lossy signal compression techniques make transmission quicker and more efficient but introduce noise when the transmitted signal is compressed. This level of noise can be controlled and restricted to an acceptable level for the vast majority of the transmission. However, there are instances where the signal data at one or more particulars levels within the bandwidth are more important than others. Alternatively, noise at particular levels of the bandwidth has a more detrimental effect than it would at other levels of the bandwidth. The aggressiveness of the compression technique can be set so that the noise in these critical sections is acceptable, but then majority of the bandwidth is only lightly compressed and the data size remains large. Keeping the data size large tends to defeat the purpose of compressing the signal in the first place.
JPEG (Joint Photographic Experts Group) compression of contone image data is one example of a lossy signal compression technique. The noise induced by JPEG compression in particular sections of the bandwidth can cause particularly visible artifacts in the decompressed image. Because of its relevance to the present invention, the detailed description is directed to localized noise reduction in the compression and decompression of an image file. However, it will be appreciated that this is purely illustrative and the invention encompasses other types of signal transmission.
JPEG compression of image data uses one of a suite of standard algorithms to reduce data size for faster transmission and more efficient storage. The quality of the resultant image is determined by the level of compression. An aggressive compression greatly reduces the file size but introduces high levels of noise. Light compression reduces the noise but the data size remains relatively large. Therefore, the optimum level of compression is a trade off between image quality and data size, having regard to the characteristics of the output device (printer or monitor), processing capabilities and resolution requirements.
During JPEG compression, the image is analyzed in blocks of 8×8 pixels. Depending on the level of compression selected, the detail in each of the blocks is reduced. In more aggressive compressions, the 8×8 blocks can become visible in the final image. The compression should be at a level where the noise in the resulting image is imperceptible. Unfortunately, there are often certain components of an image that are far more prone to decompression artifacts than the rest of the image. In these cases, the noise is imperceptible for the majority of the image, but produces artifacts in certain parts.
The noise prone areas are hard edges between strongly contrasting colors such as text on a white background.
Accordingly the present invention provides a method of preserving signal data at a predetermined level within the bandwidth of an input signal to be processed for use by an output device, the method comprising:
resealing the signal data in levels other than the predetermined level to move at least some of the signal data out of a buffer section of the bandwidth adjacent to the predetermined level;
processing the signal for use by the output device;
re-assigning any data in the buffer section to the predetermined level; and,
resealing the signal data in levels other than the predetermined level to move data back into the buffer section.
Inaccuracies in the signal processing shifts some signal data from its original level in the input signal to different level in the output signal. This shift in data generates the noise in the output signal. If data does shift because of the signal processing, there is a high probability that it only shifts to a nearby level in the bandwidth. If the signal data at a particular level is of greater importance relative to most of the other levels, the invention allows this data to be preserved at its original level with very little, if any, lost to noise.
By resealing the input signal, most or all of the data in levels near the important level can be shifted away. This effectively creates a buffer on one or both sides of the important level which quarantines the data in this level from the rest of the signal data. Any noise induced in the data from the important level is (highly likely to be) confined to the empty buffer, and so can be easily corrected. By re-mapping all data in the buffer back to the important level, the induced noise is removed. Once the data in the buffer has been mapped back to the important level, the initial resealing of the input signal can be reversed to distribute signal data across the full bandwidth.
Optionally, the input signal is image data for a color plane of an image and the output device is a printer. In these embodiments, the signal data is pixel intensity values for the color plane quantized into a number of discrete intensity levels, such that the number of levels is the bandwidth of the input signal. In a further preferred form of these embodiments, the predetermined level is the intensity level corresponding to ‘white’ (or zero color intensity). Optionally, the method preserves the data in a second predetermined level, the second predetermined level being the maximum intensity level in the bandwidth.
Preferably, the processing of the signal involves lossy transmission of the signal data. In a further preferred form, the processing of the signal involves the lossy compression of the signal data. In a still further preferred from, the processing of the signal involves lossy image compression. In a particularly preferred form, the processing of the signal includes JPEG compression.
Optionally, the input signal is resealed by quantizing the image data into a lesser number of the discrete intensity levels except for image data in the or each predetermined level. In a preferred form the image data is resealed to floating point values and then rounded to the closest of the intensity levels.
In a particularly preferred form, the method further comprises converting the image data from a first color space to a second color space wherein the resealing of the input image data is performed simultaneously with the color space conversion. In this form, the second color space is the printer color space. Optionally, the step of resealing the image data back into the or each buffer section after JPEG compression is performed via a dither matrix by adjusting the threshold values in the dither matrix.
Preferably, the image includes text characters. Preferably, the image includes line art. Preferably the image has a white background. Optionally, the color intensity values are 8-bit values and the bandwidth of the input signal is 256 levels. Optionally the predetermined level is ‘0’. Optionally, the second predetermined level is ‘255’. Optionally, the buffer section corresponding to level 0 is levels 1 to 16. Optionally, the buffer section corresponding to level 255 is levels 240 to 254.
The invention will now be described by way of example only, with reference to the embodiments shown in the accompanying drawings, in which:
As discussed in the Background to Invention,
When the image of
It will also be appreciated that the buffers need not be the same size or symmetrically positioned in the signal bandwidth. It should also be noted that the buffers need not be at the extremes of the bandwidth. For example, if the signal is image data in a YCC-style color space (luminance, chroma red and chroma blue), the important data in the chroma channels is the neutral level in the middle of the bandwidth (level 128 in 8 bit color values). The important data in the luminance channel is at the extremes.
The image of
In these areas, all pixels that were level 0 in the input image should be preserved as level 0 pixels in the output image in order to avoid the visible ringing. Likewise level 255 pixels in the input image should be kept at that level in the output image although this is usually a less visible artifact.
The quality of the JPEG compression is known, and so the section on the bandwidth in which the majority of noise induced in the level 0 pixels 12 is also known. These buffer sections 20 and 22 of the bandwidth are chosen as levels 1-16 and levels 240-254. For the purposes of this example, the vast majority of noise induced by decompressing levels 0 and 255 will appear in these buffer sections.
L
r
=L
1
+L
i.(Lh−L1)/((Lmax−1)−Lmin) eq.1
Where:
In the example shown in
L
r=16+0.878.Li
Lr can be rounded to the nearest integer or left as a floating point value to more accurately invert the resealing process when the input data 24 is distributed back across the full bandwidth.
Once the noise in levels 0 and 255 has been corrected, the rest of the data 24 can be redistributed back into the now vacant buffer sections 20 and 22, as shown in
It should also be noted that the manipulation of the image data can be achieved in a computationally efficient way by incorporating it into existing data processing steps. The resealing of the signal to move data out of the buffer sections can be done when the image data is color space converted from RGB to CMY(K). Furthermore, the re-distribution of data back across the full bandwidth can be done by adjusting the threshold values used in the dither matrix during halftoning. The Applicant's co-pending USSN (KIP001US) incorporated herein by reference describes how the histogram can be expanded (or contracted) using the dither matrix. In light of this, the computational cost for preserving the data at one or more levels in the bandwidth is relatively little, yet the tangible image quality improvement is significant. The threshold values in the compressed range in the secondary matrix are determined using Equation 2:
T
new
=L
1
+T
old.(Lh−L1)/256 Eq. 2
where:
Tnew is the compressed threshold values in the secondary dither matrix; and,
Told is the threshold value in the primary dither matrix.
The adjusted dither matrix will not affect data at the extreme levels of the bandwidth as level 0 will still be below the lowest threshold value and level 255 will still be above the highest threshold level. Only data in the levels between the buffer sections will be ‘expanded’ by the adjusted dither matrix.
As discussed above, the noise in the white background of a hard edge is more visible than the noise in the full color area along the edge. This is particularly true if the output device is a printer. In light of this, only the data in level 0 can be preserved for a significant reduction in noise. This reduces the resealing of the remaining signal so that when it is expanded back to the full bandwidth, less levels are lost and the contone image is not as coarse.
In other embodiments, the input signal is rescaled in a manner different to that the technique set out in Equation 1.
The invention has been described herein by way of example only. Skilled workers in this field will readily recognize many variations and modification that do not depart from the spirit and scope of the broad inventive concept.
