Image analysis – Image compression or coding – Pyramid – hierarchy – or tree structure
Reexamination Certificate
2000-02-04
2003-03-11
Couso, Jose L. (Department: 2721)
Image analysis
Image compression or coding
Pyramid, hierarchy, or tree structure
Reexamination Certificate
active
06532308
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This application relates generally to image signal processing, and more particularly to processing image signals in a quality priority operational mode.
2. Description of the Related Art
Typical image compression systems exploit the fact that images stored in the sample domain are almost always undermodulated (or highly autocorrelated) and therefore require significantly more storage than their information content actually requires. These systems have been unable to guarantee a worst case local error, or even a global average error in magnitude resolution at frequency, frequency response, DC accuracy, or any other hard quantification.
There are several methods of specifying a desired image accuracy in the sample domain, including exact n bit, absolute error bound, and peak signal to noise ratio. Each method assumes that the frequency response is unity over the useful passband, and that frequency resolution grows in accordance with sampling theory (3 dB/octive/dimension).
If an image is captured at a very high resolution per pixel, and quantified for storage in the sample domain at some arbitrary resolution, noise must be added at that resolution to linearize the quantification function to assure that adequate low frequency resolution is available. This noise has two undesirable effects. It reduces the resolution at Nyquist frequency to half that implied by the sample resolution, and it introduces noise information into the image which often exceeds the total image information (without the noise).
In the sample domain, the sample resolution is directly proportional to the data throughout rate. Thus, whatever equipment is available at a given cost directly dictates the available resolution. Sample resolution in both frequency and fidelity are therefore typically picked to be just adequate for an application.
Alternative approaches introduce frequency or phase uncertainty. The detailed analysis required in these domains precludes any reasonable quality priority encoding implementation. Other alternatives provide good performance in the right circumstances, but do not perform adequately in certain situations such as when quantification is applied. Vector quantification and fractal transforms are examples that generate phase and frequency domain artifacts which can be extremely difficult or impossible to bound. The RMS error of the vector can in some cases be controlled, but typically the local error bound is not controlled, nor is the worst case phase and frequency artifacts emerging from various two or three dimensional patterns of vectors.
Nonlinear phase transforms and many types of predictive error encoding systems can produce so much phase uncertainty that they are not practical for quality priority encoding. Finally, partial image transforms such as the commonly used 8×8 Discrete Cosine Transform convert any quantification error into localized anomalies in both the spatial and frequency domain, making them unsuitable for quality priority encoding, particularly when quantification is present.
Thus, there remains a need for compressed image signal processing that can operate in a quality priority mode.
SUMMARY OF THE INVENTION
The present invention includes apparatuses and methods for quality priority processing of image signals. Quality can be measured according to the modes of accuracy typically used in the sample domain, and can be guaranteed even as the image signal is processed in the encoded domain. Thus, the typical under-modulation of images is exploited to reduce bandwidth (storage, transmission, or otherwise) while the classical artifacts of image compression are avoided.
By using the quality priority image processing scheme of the present invention, an image can be captured at a very high resolution and stored without dither noise, because the desired low frequency resolutions are maintained directly to meet the requirements of sampling theory. This offers more efficient storage to assure image quality than the sample domain, which is dependent upon the presence, type, and frequency distribution of the linearizing noise. Further, the scheme allows compression without artifacts as only the information in the image is stored, in contrast to sample domain processing where redundant data is retained to ensure integrity.
In one embodiment, the quality priority image processing scheme receives a desired quality level and determines image processing settings for transformation, quantification, and encoding that ensure maintenance of the desired quality level. An input image signal is preferably subdivided in space or time to produce a plurality of subband regions. With quantification, each of these regions has a number of symbols (i.e., the symbol count), with each symbol being represented by a number of bits (i.e., the data size). An array of the data content for each region is determined as the symbol count times the data size for each region. The total data content equals a summation of the entries in the regional data content array.
The data is also encoded after transformation and quantification. While this encoding produces substantial reductions in the overall data content, at maximum entropy the data content for a given symbol can exceed the original data size. This possible excess data content can be referred to as the encoding. overhead. The data content at maximum entropy can be determined based upon the encoding overhead and the total data content. This maximum entropy data content and the desired image processing rate (e.g., images per second) are used to determine the peak data rate required to guarantee the selected quality level.
REFERENCES:
patent: 5394473 (1995-02-01), Davidson
patent: 5966465 (1999-10-01), Keith et al.
patent: 5990955 (1999-11-01), Koz
Bromberg & Sunstein LLP
Couso Jose L.
QuVIS, Inc.
LandOfFree
Quality priority image storage and communication does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Quality priority image storage and communication, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Quality priority image storage and communication will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3001971