Image analysis – Image compression or coding
Reexamination Certificate
2000-08-25
2004-06-08
Wu, Jingge (Department: 2623)
Image analysis
Image compression or coding
C375S240230
Reexamination Certificate
active
06748113
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to technique of decoding a coded image, and more particularly relates to technique of detecting and eliminating coding noise derived from encoding.
Recently, the MPEG (Moving Picture Experts Group) standards are widely used to code a picture with high compression efficiency in the fields of broadcasting, communications and storage. According to the MPEG, an image is encoded by removing spatial and temporal redundancy.
In order to remove spatial redundancy, discrete cosine transform (hereinafter referred to as DCT) and quantization are employed. First, an image is divided into units each designated as a block including 8×8 pixels, each block is converted through the DCT into coefficients of a frequency domain (hereinafter referred to as DCT coefficients), and the DCT coefficients are subjected to the quantization.
The quantization is an operation for dividing a DCT coefficient by both a quantizer matrix having a value corresponding to each frequency of a DCT region and a quantizer scale. Through the quantization, a value of a frequency component with a small DCT coefficient becomes zero. In general, energy is collected in a low frequency region in an image signal, and hence, a high frequency component is eliminated through this operation. However, the visual characteristic of a human being is poorer in a higher frequency region, and therefore, degradation in picture quality is less conspicuous as the quantizer scale used in the quantization is smaller (i.e., a step in the quantization is smaller).
In order to remove temporal redundancy, motion compensation is employed. In the motion compensation, an area closest to a macroblock in coding within a reference frame is selected with a macroblock of 16×16 pixels used as a unit, so as to encode difference values from the reference frame. When motion is not very fast, the difference values are substantially zero, and thus, temporal redundancy can be removed.
In general, when the bit rate of a code is high (i.e., the compression rate is low), the degradation in picture quality is very inconspicuous in the MPEG. When the bit rate is lower (i.e., the compression rate is higher), however, coding noise becomes more conspicuous and the picture quality is degraded. Typical coding noise occurring in the MPEG are blocking artifact (sometimes designated as block distortion or block noise) and ringing artifact (sometimes designated as ringing noise, corona noise or mosquito noise).
Blocking artifact is a phenomenon where a block boundary can be obviously seen in a tile-like shape. This noise is caused when an image signal within a block has merely low frequency components and the frequency component values are different from those of an adjacent block.
Ringing artifact is noise occurring in the vicinity of the edge of an image like a flying mosquito. This noise is caused because a high frequency component inherent to an image signal has been eliminated through the quantization.
Blocking artifact and ringing artifact are greatly remarkable types of degradation in picture quality unlike analog noises, and hence, several methods have been proposed for eliminating these coding noises.
An example of the method of eliminating blocking artifact is disclosed in Japanese Laid-Open Patent Publication No. 5-308623 (first conventional method). In this conventional method, a filter whose frequency characteristic is determined depending upon the maximum frequency of DCT coefficients and motion activity obtained based on a motion vector is applied.
Also, Japanese Laid-Open Patent Publication No. 7-177521 discloses a method in which a filter characteristic is determined by a quantizer scale and the size of a motion vector so that blocking artifact can be eliminated by applying a filter with the determined characteristic to a block boundary of a decoded image (second conventional method). In this conventional method, in the case where the quantizer scale is large and the motion vector is small, the filter characteristic is determined based on the quantizer scale alone. Alternatively, in the case where the quantizer scale is small and the motion vector is large, the filter characteristic is determined based on the size of the motion vector alone. Furthermore, in the case where both the quantizer scale and the size of the motion vector are medium, the filter characteristic is determined based on both of them.
On the other hand, an example of the method of eliminating ringing artifact is disclosed in Japanese Laid-Open Patent Publication No. 6-311499 (third conventional method). In this conventional method, an absolute sum of specific DCT coefficients of each block is obtained, so as to change the threshold value of a filter in accordance with the obtained absolute sum. Furthermore, in a frame that has been inter-frame coded, elimination of ringing artifact is carried out on a difference image, and then, reference frames where ringing artifact has been eliminated are added to the difference image so as to obtain a decoded image.
In general, occurrence of blocking artifact depends not on the maximum frequency of DCT coefficients but on the frequency component distribution of DCT coefficients and a difference in the frequency component distribution of DCT coefficients between adjacent blocks. When the filter characteristic is determined as in the first conventional method by using the high frequency component alone of the DCT coefficients, capability of eliminating blocking artifact is lowered.
The first conventional method further has the following problems: Since the filter characteristic is determined individually in the horizontal direction and the vertical direction, the operation required for the filter determination becomes numerous in volume. Furthermore, when the filter characteristic is determined based on DCT coefficients, capability of detecting blocking artifact is lowered in employing inter-frame coding (non-intra coding). In addition, in the case where the operation is controlled so that a strong filter is applied when the motion activity is large, no consideration is paid to the property of the reference frame. For example, in the case where motion compensation is accurately carried out even though the motion activity is large, a filter is still applied even if the pixel values in the difference image are all zero. As a result, the image can be degraded in the quality, for example, blurred. Accordingly, this method degrades the picture quality when applied to a non-intra coded image.
Moreover, even when the motion vector is large, a decoded image may include a high frequency component. Furthermore, with respect to a frame that has been inter-frame coded, even when the quantizer scale is large, a reference frame may include a high frequency component. In such a case, a decoded image may include a high frequency component. In this case, when a filter is applied to the decoded image as in the second conventional method, the picture quality is degraded.
Furthermore, the third conventional method has the following problems: Since the absolute sum of specific DCT coefficients of each block is obtained, calculation of the absolute sums of the DCT coefficients requires numerous processing. Moreover, since a decoded image is obtained by adding reference frames where ringing artifact has been eliminated, reference frames used in decoding are different from those used in coding. Accordingly, errors are accumulated on the decoded image, which makes the degradation in the picture quality conspicuous.
SUMMARY OF THE INVENTION
The present invention was devised to overcome the aforementioned conventional problems. An object of the present invention is providing a noise detecting method and a noise detector capable of definitely detecting a block boundary where blocking artifact occurs or a block where ringing artifact occurs so as to accurately detect such coding noise even in a non-intra coded block, and providing an image decoding apparatus capable of definitely eliminating coding noise with deg
Iwasaki Eiji
Kondo Satoshi
Takahashi Hideya
Taniuchi Hiroshi
Matsushita Electric Insdustrial Co., Ltd.
Nixon & Peabody LLP
Wu Jingge
LandOfFree
Noise detecting method, noise detector and image decoding... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Noise detecting method, noise detector and image decoding..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Noise detecting method, noise detector and image decoding... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3364457