Patent Grant
7835044
Embodiments are related to data-processing methods and systems. Embodiments are also related to image-processing devices and techniques. Embodiments are additionally related to color space transformation in image-processing applications.
In color space transformations based on three-dimensional (3D) lookup with tetrahedral interpolation, the interpolation stage calculates the estimated [CMYK] output value based on the given CMYK values of the nearest four nodes to the current location in the input 3D space. Such a methodology assumes that the color transformation is relatively smooth and that the nodes are placed close enough such that the transformation is approximately piece-wise linear between adjacent nodes.
The approximation level (and the resulting color transformation accuracy) is generally dependent on the number of nodes, their relative positions, and the non-linearity of the color transformation in a given neighborhood. The interpolation computation equation in accordance with one existing method can be described as follows:
Outputi=1/128(128*P0i+Xfrc*ΔXi+Yfrc*ΔYi+Zfrc*ΔZi)
The above equation assumes that the CMYK node difference values (ΔX, ΔY, ΔZ) are represented as signed 8-bit numbers, and the fractional position values (Xfrc, Yfrc, Zfrc) within a given tetrahedron are represented as unsigned 8-bit numbers after the subsequent rank ordering described below.
The above equation can be computed per output color component (C, M, Y, and K), where the node delta values (ΔX, ΔY, ΔZ) are dependent on the tetrahedron containing the mapped input point in the input color space (e.g., in (L, a, b) or (Y, Cb, Cr)).
Note that, in general, the chrominance values (a, b) or (Cb, Cr) are signed numbers while the luminance values (L or Y) are unsigned numbers.
In one application, each cube of the input color space can be divided into six non-overlapping tetrahedrons of the same size. Alternatively, the space can be divided into five tetrahedrons (i.e., a fewer number), but in this case the center tetrahedron has a larger size and this somewhat complicates the calculations. Thus, the focus can be on the six tetrahedron case. Table 1 below illustrates the node value delta computation for all possible tetrahedrons with reference to the existing method.
The specific tetrahedron in which the input location resides can be easily determined by simply ordering the fractional delta values (Xfrc, Yfrc, Zfrc) in rank order. The interpolation algorithm computes the output value using rounding and limiting the final result to the preferred output range (e.g., between 0 and 255 for 8-bit output). After the interpolated CMYK point has been calculated, each of its color components can be used as an address to four different 1D lookup tables. These lookup tables provide a means to linearize the main CMYK color interpolation table of node values with respect to the individual {C, M, Y, K} color response. In addition, the 1D lookup tables can be used to compensate for changes in the individual {C, M, Y, K} tone reproduction curve as the machine drifts over time and/or the level of toner/ink is depleted without having to re-calibrate the node positions of the main color tables. The values accessed from the four lookup tables represent the adjusted CMYK values that comprise the final output value.
The entire color interpolation method in accordance with one existing method can then be described as following: for each output channel (C, M, Y, and K), four output values representing the nearest nodes to the mapped Lab or YCC input point are looked-up in a table and used in a tetrahedral interpolation step to find the approximated CMYK output value. The resulting output value is then adjusted by a byte-to-byte mapping using four 1-D lookup tables.
Color space transformation via 3D Lookup table with tetrahedral interpolation is a well-known algorithm for converting from one color space (e.g., RGB or Lab) to another (e.g., CMYK for printing). In software implementations on general purpose processors, for example, the computational steps are processed in sequential fashion, one color separation at a time. Although the fractional position within a specific tetrahedron is the same for each color separation, the node values are different, thus requiring individual calculations per color separation. In contrast, the algorithm can be efficiently implemented in hardware by packing the data and processing the transformation of the color channels in parallel. By taking advantage of this property, it can be demonstrated that certain media processors with data-level parallelism can deliver higher-performance color space transformation relative to general purpose processors and common hardware implementations.
One known prior art application involves interpolation techniques for improved efficiency and speed in performing color space conversions. In such a case, a radial interpolation technique can accomplish an interpolation by generating successive subcubes. A value of a vertex of the final subcube generated can be used as the result of the interpolation. Subcubes are generated by averaging a selected vertex value with the vertex values of each of the remaining vertices. A pruned radial interpolation technique employs a subset of the vertex values of the initially selected cube to generate the result of the interpolation, thereby improving upon the efficiency of the radial interpolation. A tetrahedral interpolation technique accomplishes an interpolation by generating successive subcubes.
A value of a vertex of the final subcube generated can be used as the result of the interpolation. Subcubes are generated by applying a mathematical relationship which allows computation of subcube vertex values through a series of logical AND, logical OR and averaging operations. A pruned tetrahedral interpolation technique employs a subset of the vertex values of the initially selected cube to generate the result of the interpolation, thereby improving upon the efficiency of the tetrahedral interpolation. A common hardware implementation of pruned radial interpolation and pruned tetrahedral interpolation uses the common hardware structure of the two techniques with multiplexing of the input vertex values to allow performance of either a pruned radial interpolation or a pruned tetrahedral interpolation. Non-symmetric pruned radial and Non-symmetric pruned tetrahedral interpolation permit interpolation using interpolation data values distributed throughout the color space with a resolution that varies according to characteristics of the color space. Multiplexing of the interpolation data values to the non-symmetric pruned radial interpolation hardware and to the non-symmetric pruned tetrahedral interpolation hardware allows for a common hardware implementation.
Another prior art technique involves tetrahedral interpolation by rewriting the interpolation in terms of ordered differentials and color differences to lower the computational complexity.
One of the problems with these prior art techniques is that such methodologies and related devices are inefficient and cannot be easily adapted to a wide range of image processing applications. In order to overcome these problems, it is believed that the embodiments disclosed herein can be implemented for increased efficiency in both processing and hardware applications.
The following summary is provided to facilitate an understanding of some of the innovative features unique to the embodiments disclosed and is not intended to be a full description. A full appreciation of the various aspects of the embodiments can be gained by taking the entire specification, claims, drawings, and abstract as a whole.
It is, therefore, one aspect of the present invention to provide for improved data-processing systems and methods.
It is another aspect of the present invention to provide for improved image-processing systems and methods.
It is a further aspect of the present invention to provide for an efficient method for tetrahedral interpolation computations using data-level parallelism by taking advantage of media processor parallel instructions while providing a more efficient calculation of the approximated output point by delivering higher-performance color space transformation in image-processing applications
The aforementioned aspects and other objectives and advantages can now be achieved as described herein. A system and method for tetrahedral interpolation computations using data-level parallelism that takes advantage of data-level parallelism in media processors is claimed. If the tetrahedron points in a 3D lookup table are packed together in a memory, the interpolation computation can be implemented without extra instructions to unpack them. An algebraic manipulation of the interpolation equation allows computing the difference on the fraction coefficients instead of the tetrahedron node values. Not only will this technique preserve the full precision without over or underflow, but the packed data from the 3D lookup can be used directly as is (without unpacking), thereby allowing a faster implementation of the color space transformation overall and implementing as part of a direct-copy image path on a media processor. Such systems and methods allow the implementation of the direct copy pipeline to function at the required performance rate as requested by a customer specification while obtaining the required product design speed.
Features, aspects and advantages of the present invention will become better understood with regard to the following description, appended claims, and accompany drawings where:
a-3e illustrates interpolation volumes in accordance with preferred embodiments
The particular values and configurations discussed in these non-limiting examples can be varied and are cited merely to illustrate the aspects of the present invention and are not intended to limit the scope of the invention.
Referring to
Referring to
Based on the foregoing, it can be appreciated that by taking advantage of data-level parallelism in media processors (e.g., such as the INTEL MXP5800), i.e. if the tetrahedron points (usually 8-bit values per color channel) in the 3D lookup table are packed together in memory (as one 32-bit word, say), the interpolation computation can be implemented without extra instructions to unpack them. An algebraic manipulation of the interpolation equation allows computing the difference on the fraction coefficients instead of the tetrahedron node values.
Not only will the inventive technique described herein preserve the full precision without over or underflow, but the packed data from the 3D lookup can be used directly as is (without unpacking), thereby allowing a faster implementation of the color space transformation overall. This method has been implemented as part of a direct copy image path on a media processor. This method allows the implementation of the direct copy pipeline to work at the required performance rate as requested by the customer specifications. Without this method, the color space transformation performance would have failed to meet the required product design speed.
In order to take advantage of the media processor parallel instructions, and to make a more efficient calculation of the approximated output point, the algorithm can be slightly modified. For cost/performance reason, most media processors use a fixed HW precision arithmetic (8-bits, for example). Since the node values are 8-bit unsigned values, the result of the subtraction operations in the existing equation with reference to Table 1, generally requires 9 bits to preserve the full precision.
Thus a general loss of precision might occur when the existing equation is used as is. However, by re-organizing the existing equation in matrix form, there is an opportunity to take the subtraction on the fractional values rather than the tetrahedron node values. Since the fractional values are limited to 7-bit unsigned values, the result of a subtraction operation in this preferred method requires 8 bits of precision and thus can be implemented within the existing HW precision without any loss of accuracy. This makes it particularly appealing for use of dual instructions for the multiply-accumulate operations associated with the interpolation.
Rank-ordering the fractional values can be used to identify the specific tetrahedron which contains the mapped input point. The rank-ordering process can assure that the delta of the fractions is always a positive result due to the subtraction of a small number from a large number. This applies for any tetrahedron case out of the six possible tetrahedrons. Table 2 below shows the fractional delta values for all possible tetrahedrons. In the end, this leads to the use of unsigned dual (or quad) instructions in the computation of the interpolation output. Thus, the implementation of the tetrahedral interpolation stage in media processors with data-level parallelism can be efficiently computed following the definition of the last term in the modified formed equation and the delta values defined in Table 2 below.
This summarizes the sequence of steps that can be applied for all tetrahedrons. After using the fractional values to identify the tetrahedron that contains the mapped input point, the fractional delta values can be computed according to Table 2 below. Then the tetrahedron node values pairs CM and YK (if 16-bit words are packed together) or CMYK quad node values (if the packing is done on 32-bit words) can be retrieved from the 3D lookup table and used directly in the unsigned-dual (or quad) multiply-accumulate instructions that compute the interpolation outputs. Once the interpolation of the CMYK channels has been calculated, the interpolation outputs are packed together (as an outcome of the use of dual or quad instructions) and can be directly adjusted using a byte-to-byte mapping to implement additional 1D CMYK lookup tables to provide tone reproduction curves. These 1D lookup tables provide a means to linearize the main CMYK color interpolation table of node values with respect to the individual {C, M, Y, K} color response. In addition, the 1D lookup tables can be used to compensate for changes in the individual {C, M, Y, K} tone reproduction curve as the machine drifts over time and/or the level of toner/ink is depleted without having to re-adjust the tetrahedral node positions of the main color tables.
Table 2 below, for example illustrates the calculation of Fraction Delta Values with reference to the modified method.
The alternative Interpolation computation equation is referred with the modified method is given as follows:
Outputi=1/128[Δ0*P0i+Δ1*P1i+Δ2*P2i+Δ3*P3i]
Where iε{C, M, Y, K}, P0i, P1i, P2i, P3i are the four tetrahedron points, and Δ0, Δ1, Δ2, Δ3 are computed according to Table 2.
Note that embodiments can be implemented in the context of modules. Such modules may constitute hardware modules, such as, for example, electronic components of a computer system. Such modules may also constitute software modules. In the computer programming arts, a software module can be typically implemented as a collection of routines and data structures that performs particular tasks or implements a particular abstract data type.
Software modules generally are composed of two parts. First, a software module may list the constants, data types, variable, routines and the like that can be accessed by other modules or routines. Second, a software module can be configured as an implementation, which can be private (i.e., accessible perhaps only to the module), and that contains the source code that actually implements the routines or subroutines upon which the module is based. The term module, as utilized herein can therefore refer to software modules or implementations thereof. Such modules can be utilized separately or together to form a program product based on instruction media residing in a computer memory that can be implemented through signal-bearing media, including transmission media and recordable media, depending upon design considerations and media distribution goals. Such instruction media can thus be retrieved from the computer memory and processed via a processing unit, such as, for example, a microprocessor.
The system and/or method described above, for example, with respect to
It will be appreciated that variations of the above-disclosed and other features and functions, or alternatives thereof, may be desirably combined into many other different systems or applications. Also that various presently unforeseen or unanticipated alternatives, modifications, variations or improvements therein may be subsequently made by those skilled in the art which are also intended to be encompassed by the following claims.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5581376 | Harrington | Dec 1996 | A |
| 5751845 | Dorff et al. | May 1998 | A |
| 5751926 | Kasson et al. | May 1998 | A |
| 6031642 | Vondran, Jr. | Feb 2000 | A |
| 7483174 | Tadas | Jan 2009 | B2 |
| 20040049497 | Curry et al. | Mar 2004 | A1 |
| 20040061902 | Tang et al. | Apr 2004 | A1 |
| 20040109185 | Hung et al. | Jun 2004 | A1 |
| Number | Date | Country | |
|---|---|---|---|
| 20070247648 A1 | Oct 2007 | US |
