Image analysis – Image enhancement or restoration – Edge or contour enhancement
Reexamination Certificate
2000-05-12
2001-10-16
Rogers, Scott (Department: 2624)
Image analysis
Image enhancement or restoration
Edge or contour enhancement
C382S275000, C382S205000
Reexamination Certificate
active
06304678
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention is related to the field of image processing, and more particularly to techniques for post-processing of decoded images to remove undesirable artifacts such as ringing and blocking.
Image compression is useful for efficient electronic storage of images as well as efficient transmission of images over a communications medium, such as the Internet. Many image compression techniques are in use today. The generalized image compression scheme employs a cascade of functional elements. For example, a transform may be applied to the original image to convert the image information to a more convenient domain, such as the frequency domain. The transformed image information is quantized to discrete levels that can be represented by a finite digital word, and the quantized result is compressed according to a compression coding algorithm. At the receiver, these functions are performed in reverse order to recover a facsimile of the original image.
Many compression techniques introduce certain types of noise, referred to as “artifacts”, into the image. For example, an algorithm known as Joint Picture Expert Group (JPEG) encoding tends to introduce blocking artifacts at medium and low bit-rates. Blocking artifacts appear as subtle rectangular segmentation in the decoded image. In the JPEG algorithm, blocking arises from the use of short and non-overlapping basis functions.
Blocking can be reduced or eliminated using other transforms, such as wavelet transforms, that have overlapping basis functions. However, these may introduce spurious oscillations in the vicinity of major edges at low bit-rates. Such coding artifacts are called ringing artifacts. In a wavelet encoding scheme, ringing artifacts are caused by the abrupt truncation of the high frequency wavelet coefficients. It is desirable to obtain an image that is as free of compression-related artifacts as possible, thus improving image quality at low bit rates.
An artifact-free image can be estimated from the compressed image by maximum a posteriori (MAP) estimation techniques. The problem is to generate an artifact-free estimate f′ of an original image f given a compressed image g. In MAP estimation approaches, the estimate f′ is considered to be a random variable whose properties are modeled by a probability density, following the Bayesian viewpoint. The MAP estimate is the estimate f′ that maximizes the posterior probability that is expressed in terms of a conditional probability and a prior probability. For transform-based coders-decoders (codecs), the conditional probability is modeled in the transform domain, while the prior probability is modeled in the spatial domain. This aspect significantly increases the computational complexity when the solution is obtained through iterative algorithms.
In addition to MAP techniques, there are other techniques that can be used to reduce artifacts, such as an algorithm known as Projection Onto Convex Sets (POCS). However, algorithms such as MAP and POCS suffer from the disadvantage of requiring the use of both the forward and the inverse transforms. Also, these algorithms are iterative in nature, so the forward and the inverse transforms are needed at each of several iterations of the algorithm. This aspect of existing algorithms increases their computational complexity significantly.
It would be desirable to reduce artifacts appearing in decoded compressed images while reducing the extensiveness and complexity of the required computational resources.
BRIEF SUMMARY OF THE INVENTION
In accordance with the present invention, a technique for post-processing decoded compressed images is disclosed that reduces decoding-related artifacts. A simplified version of the technique is non-iterative in nature, and provides good results without requiring extensive computational resources.
The technique employs a maximum likelihood estimation viewpoint on an estimate of an original image f. This estimate is regarded as a quantity with an unknown value. The image is modeled as a montage of “flat surfaces” of different intensities, where the number of flat surfaces and their respective intensities are generally different in different regions of the decoded image. The intensity of each pixel is conditionally adjusted to that of a corresponding flat surface in a window region surrounding the pixel. The adjustment of the pixels results in the reduction of certain types of noise in the vicinity of edges in the image, such as ringing and blocking artifacts.
In a general algorithm, the flat surface model is fitted to the observed image by estimating the model parameters. An algorithm known as the “k-means” algorithm is used in conjunction with a hierarchical clustering algorithm to achieve this fit. A cluster similarity measure (CSM) is used in the hierarchical clustering algorithm to determine the number of intensity clusters, and hence flat surfaces, that should be used to model a window region surrounding a pixel of interest. The pixel intensity is then adjusted to an estimated value which is the mean intensity of the cluster in which the pixel falls.
According to a simplified version of the above method, a three-cluster model is used in which the cluster centers are initialized by a deterministic rule. This simplified method performs a non-iterative mapping between the neighboring pixel values and the estimate, and is therefore particularly attractive for its good performance and simplicity. In one example, the disclosed methods are applied to remove ringing artifacts in images compressed by the JPEG2000 algorithm. The results show effective and efficient elimination of ringing artifacts with improved image quality.
Other aspects, features, and advantages of the present invention are disclosed in the detailed description that follows.
REFERENCES:
Fan, Zhigang et al., “Reducing Artifacts in JPEG Decompression by Segmentation and Smoothing”, Proceedings of the International Conference on Image Processing, IEEE 1996, p. 17-20, vol. 2 of 3, expecially section 2.
Yang, Seungjoon et al., “Blocking Effect Removal Using Robust Statistics and Line Processing”, 1999 IEEE Third Workshop on Multimedia Signal Process, IEEE, 1999, p. 315-320, especially Introduction.
Hu Yu Hen
Nguyen Truong Q.
Tull Damon L.
Yang Seung-joon
Rogers Scott
The Trustees of Boston University
Weingarten, Schurgin Gagnebin & Hayes LLP
LandOfFree
Image artifact reduction using maximum likelihood parameter... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Image artifact reduction using maximum likelihood parameter..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image artifact reduction using maximum likelihood parameter... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2553359