WO2006113057A2

WO2006113057A2 - Down-sampling and up-sampling processes for chroma samples

Info

Publication number: WO2006113057A2
Application number: PCT/US2006/011135
Authority: WO
Inventors: Ying M. Chen; Jiefu M. Zhai; Peng Yin; Edouard M. Francois
Original assignee: Thomson Licensing
Priority date: 2005-04-15
Filing date: 2006-03-24
Publication date: 2006-10-26
Also published as: WO2006113057A3

Abstract

A process for down-sampling and up-sampling processes for chroma samples relies on the use of multi-tap filters for sampling the horizontal and vertical chroma components. In this way, no phase shift occurs in the vertical (Y) component.

Description

DOWN-SAMPLING AND UP-SAMPLING PROCESSES FOR CHROMA SAMPLES CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority under 35 U.S. C. 119(a) to European Patent Application EP 051003017.9, filed April 15, 2005, the teachings of which are incorporated herein.

TECHNICAL FIELD

This invention relates to a technique for sampling an image to obtain color information

BACKGROUND ART

A typical digital image comprises a plurality of individual picture elements, commonly referred to as pixels. Different techniques exist for representing image information associated with each pixel, especially in connection with image encoding and decoding. For example, each pixel can be represented by separate red, green and blue color values. Alternatively, in accordance with the CCIR 601 standard, each pixel can be represented by a value for pixel luminance (Y) and separate values representing red and blue color difference components (U and V, respectively), sometimes referred to as to Cb and Cr, respectively. (Knowing the U and V color difference components permits derivation of the green color different component Cg).

Establishing the Y, U and V values for each pixel occurs through a process known as "sampling." A typical Y, U and V sample for a given pixel appears as a three-part ratio, with the value of pixel luminance (Y) being 4. Thus, the process of sampling the Y, U and V values for each pixel has become known as 4:4:4 sampling.

The human eye has a lower sensitivity to chroma (color) as compared to luminance (brightness). Thus, sampling the chroma information less frequently than the luminance still affords good image quality while reducing memory requirements. A common sampling technique, referred to as YUV 4:2:0 or simply 4:2:0, samples the luminance for each pixel, but only collects the average of each pair of successively adjacent Cb and Cr values that lie in odd-number columns of pixels. The YUV 4:2:0 sampling technique inherently possess the problem of chroma component position shifting in a down-sampled sequence. Employing a different down-sampling filter in the vertical and horizontal directions could help but would introduce a phase shift in chroma component during up-sampling, which can affect the coding efficiency of some coding tools.

Thus, a need exists for a sampling technique that overcomes the aforementioned disadvantages of the prior art.

BRIEF SUMMARY OF THE INVENTION

Briefly, in accordance with a preferred embodiment, there is provided a method for sub-sampling color information in an image. In accordance with the method, the image undergoes downsampling in a vertical direction via a multi-tap filter having an even number of taps and undergoes downsampling in a horizontal direction via a multi-tap filter having an odd number of taps to yield a sub-sampled color component sample.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 graphically illustrates a set of YUV 4:2:0 sample positions associated with the MPEG 1 compression standard;

Figure 2 graphically illustrates a set of YUV 4:2:0 sample positions associated with one of the MPEG 2, MPEG 4 and H.264 coding standards;

Figure 3 graphically illustrates a set of samples obtained by horizontal down-sampling using 3 -tap filter in accordance with a first preferred embodiment of the present principles;

Figure 4 graphically illustrates a set of samples obtained by vertical down-sampling using 2-tap linear filter in accordance with a second preferred embodiment of the present principles;

Figure 5 graphically illustrates a set of samples obtained by horizontal up-sampling using 2 tap filter in accordance with a third preferred embodiment of the present principles; and

Figure 6 graphically illustrates a set of samples obtained by vertical up-sampling using linear filters in accordance with a fourth preferred embodiment of the present principles; DETAILED DESCRIPTION

FIGURE 1 depicts a set of Y UV sample positions sampled using a first type (i.e., type 1) 4:2:0 sampling for a portion of an image coded in accordance with the MPEG 1 compression standard. Each "X" in FIGURE 1 represents a luminance sample associated with a corresponding pixel in the image. Each "O" in FIGURE 1 represents a chroma sample. During 4:2:0 sampling on each line of pixels, one chroma gets stored for two luminance samples in each horizontal line and in each vertical line. In other words, each block of four pixels yields four luma samples but only one chroma sample. Thus, every other column and row yields two chroma samples, with the intervening rows yielding no samples.

FIGURE 2 depicts a set of Y UV sample positions sampled using a second type (i.e., type 2) :2:0 sampling for a portion of an image coded in accordance with one of the MPEG 2, MPEG 4 or H.264 compression standards. As with FIG. 1, each "X" in FIGURE 2 represents a luminance sample associated with a corresponding pixel in the image, whereas, each "O" in FIGURE 1 represents a chroma sample. Like type 1 sampling, type 2 sampling also yields one chroma sample for each bock of four pixels. To achieve spatial scalability with a scalable video coding scheme such as MPEG-4, a down-sampling filter (not shown) serves to down-sample low resolution images. For images encoded using the H.264 video coding standard, a 6-tap filter (not shown) serves to up-sample a reconstructed picture. The down- sampling filter used for MPEG-4 has its filtering characteristic given by the relationship

h[n]=[2, 0,-4,-3, 5, 19, 26, 19, 5,-3,-4, 0, 2]/64

which affords the filter an odd-length. The 6-tap up-sampling filter used in connection with H.264 has its filtering characteristic given by the relationship:

_g[n]= [0, 0, 1, 0, -5, 0, 20, 32, 20, 0, -5, 0, 1, 0, 0]/32.

Both filters provide both luma and chroma components. No phase shift occurs in connection with the luma (Y) component, a crucial requirement for the efficiency of some SVC modes, such as INTRA-BL. Current solutions for achieving 4:2:0 sampling, such as the filters described above, often gives rise to a phase difference between chroma samples and luma samples. Since the chroma samples do not have as high a correlation as luma samples, shorter FIR filters will not improve coding efficiency and image quality.

In accordance with the present principles, using different length filters for achieving sampling in the horizontal and vertical directions can reduce the problem of phase shifting that occurs when performing 4:2:0 image sampling. Use of a 2-tap filter having a filter characteristic of f[n]=[l,l]/2 offers a possible solution for down-sampling both in the horizontal or vertical directions but does not necessarily offer a solution to reducing phase differences. A reduction in the phase difference can occur using a 3 tap filter [1, 2, l]/4, shown in FIGURE 3 to sample in the horizontal direction, whereas a 2 tap filter [3, l]/4, which is actually a linear filter as shown in FIGURE 4, samples in the vertical direction. Thus using an odd length filter in the horizontal direction and even length filter in the vertical direction will reduce phase differences.

Another solution to reducing chroma sample phase differences is to make use of bilinear interpolation, such as is used in H.264/AVC up-sampling for half/quarter pixel motion compensation shown in FIG. 4. Yet another solution is using 2 tap filters for the horizontal and vertical up-sampling processes. However the weights of the filters based on the distances between the locations of original samples (marked by dashed lines in FIG. 5 and dashed and solid lines in Figure 6) and that of the respected up-sampled (indicated by solid lines Figure 5 and dashed lines in Figure 6). Vertically, for the odd pixels, a linear filter of [1, 7]/8 is used and for even pixels, a linear filter of [5, 3]/8 is used.

The above-described down-sampling and up-sampling techniques are only for chroma components, for luma components, MPEG-4 down-sampling filter and H.264 6-tap filter are used.

The foregoing described a technique for sampling color components in a digitized image.

Claims

CLAMS

1. A method for sub-sampling color information in an image, comprising the steps of: _ downsampling the image in a vertical direction via a first multi-tap filter having an even number of taps and downsampling the image in a horizontal direction via a second multi-tap filter having an odd number of taps to yield a color component sample.

2. The method according to claim 1 wherein the step of downsampling the image in the vertical direction comprises the step of downsampling using a two-tap filter.

3. The method according to claim 1 wherein the step of downsampling the image in the vertical direction comprises the step of downsampling using a two-tap linear filter

3. The method according to claim 1 wherein the step of downsampling the image in the horizontal direction comprises the step of downsampling using a three-tap filter.

4. The method according to claim 1 further comprising the step of upsampling the image in at least one of a horizontal and vertical directions using bi-linear interpolation.

5. The method according to claim 1 further comprising the step of upsampling the image in at least one of a horizontal and vertical directions using an even length multi-tap filter.

6. The method according to claim 1 further comprising the step of upsampling the image in the vertical direction using first and second linear filters for odd and even pixels, respectively.

7. A method for sub-sampling color information in an image, comprising the steps of: downsampling the image in a vertical direction via a first multi-tap filter having an even number of taps to yield vertical sub-sampled color samples; downsampling the image in a horizontal direction via a second multi-tap filter having an odd number of taps to yield horizontal sub-sampled color samples; and upsampling the image in at least one of a horizontal and vertical directions using one of bi-linear interpolation or a multi-tap even length filter.